# Harvey's Legal Agent Benchmark v1 / criteria pass rate white collar defense investigations > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/vals-ai%3Ahlab-v1%3Acriteria_pass_rate__white_collar_defense_investigations. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/vals-ai%3Ahlab-v1%3Acriteria_pass_rate__white_collar_defense_investigations. Tests an agent's ability to complete legal work using documents, spreadsheets, presentations, and file-system tools Canonical URL: https://benchize.com/benchmarks/vals-ai%3Ahlab-v1%3Acriteria_pass_rate__white_collar_defense_investigations Last updated: Wed, 21 Oct 2026 18:52:20 GMT Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations. Tests an agent's ability to complete legal work using documents, spreadsheets, presentations, and file-system tools Professional work · Models tested: 49 · Score publishers: Vals AI ## Model results | # | Model | Score | ± | Cost | Latency | | | --- | --- | --- | --- | --- | --- | --- | | 1 | [Muse Spark 1.3](https://benchize.com/models/muse-spark-1-3-xhigh.md)Extra high | 94.28% | SE ±1.35 pp | $2.94 | 1,094.13source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 94.276 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.939674 \| USD \| per_test \| Lower is better \| \| `latency` \| 1094.13 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 131072 provider Meta reasoning Not reported Reasoning effort xhigh Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.35 pp sourcePopulation 81 stderr 1.35 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label meta/muse_spark_1_3 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/meta~1muse_spark_1_3` | | | | | | | | 2 | [Gemini 4 Argon](https://benchize.com/models/gemini-4-argon-high.md)High | 93.94% | SE ±1.39 pp | $10.94 | 2,215.2source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 93.939 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 10.940583 \| USD \| per_test \| Lower is better \| \| `latency` \| 2215.2 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 262144 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.39 pp sourcePopulation 81 stderr 1.387 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-4-argon Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-4-argon` | | | | | | | | 3 | [Muse Spark 1.2](https://benchize.com/models/muse-spark-1-2-xhigh.md)Extra high | 93.27% | SE ±1.46 pp | $1.58 | 1,278.39source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 93.266 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 1.579724 \| USD \| per_test \| Lower is better \| \| `latency` \| 1278.39 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 131072 provider Meta reasoning Not reported Reasoning effort xhigh Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.46 pp sourcePopulation 81 stderr 1.457 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label meta/muse_spark_1_2 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/meta~1muse_spark_1_2` | | | | | | | | 4 | [Muse Wonder](https://benchize.com/models/muse-wonder-max.md)Max | 92.93% | SE ±1.49 pp | $6.12 | 1,128.13source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 92.929 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 6.121717 \| USD \| per_test \| Lower is better \| \| `latency` \| 1128.134 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 131072 provider Meta reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.49 pp sourcePopulation 81 stderr 1.49 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label meta/muse-wonder Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/meta~1muse-wonder` | | | | | | | | 5 | [MiMo-V2.6-Pro](https://benchize.com/models/mimo-v2-6-pro.md) | 89.56% | SE ±1.78 pp | $0.17 | 1,020.6source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 89.562 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.16563 \| USD \| per_test \| Lower is better \| \| `latency` \| 1020.596 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider Xiaomi reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.78 pp sourcePopulation 81 stderr 1.777 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label xiaomi/mimo-v2.6-pro Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/xiaomi~1mimo-v2.6-pro` | | | | | | | | 5 | [Muse Spark 1.1](https://benchize.com/models/muse-spark-1-1-xhigh.md)Extra high | 89.56% | SE ±1.78 pp | $0.74 | 532.96source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 89.562 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.739764 \| USD \| per_test \| Lower is better \| \| `latency` \| 532.962 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 256000 provider Meta reasoning Not reported Reasoning effort xhigh Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.78 pp sourcePopulation 81 stderr 1.777 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label meta/muse_spark_1_1 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/meta~1muse_spark_1_1` | | | | | | | | 7 | [DeepSeek V4.1 Flash 1017](https://benchize.com/models/deepseek-v4-1-flash-1017-max.md)Max | 89.23% | SE ±1.8 pp | $0.63 | 1,260.42source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 89.226 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.628043 \| USD \| per_test \| Lower is better \| \| `latency` \| 1260.418 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 384000 provider DeepSeek reasoning Not reported Reasoning effort max Temperature 1 topP 1 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.8 pp sourcePopulation 81 stderr 1.802 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label deepseek/deepseek-v4.1-flash-1017 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/deepseek~1deepseek-v4.1-flash-1017` | | | | | | | | 8 | [MiniMax M3](https://benchize.com/models/minimax-m3.md) | 88.55% | SE ±1.85 pp | $1.12 | 1,106.79source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 88.552 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 1.117058 \| USD \| per_test \| Lower is better \| \| `latency` \| 1106.794 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 512000 provider MiniMax reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.85 pp sourcePopulation 81 stderr 1.851 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label minimax/MiniMax-M3 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/minimax~1MiniMax-M3` | | | | | | | | 9 | [Gemini 3.8 Flash](https://benchize.com/models/gemini-3-8-flash-high.md)High | 87.88% | SE ±1.9 pp | $3.12 | 2,041.39source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 87.879 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 3.121405 \| USD \| per_test \| Lower is better \| \| `latency` \| 2041.394 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.9 pp sourcePopulation 81 stderr 1.897 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.8-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.8-flash` | | | | | | | | 10 | [Claude Opus 4.7](https://benchize.com/models/claude-opus-4-7-max.md)Max | 87.54% | SE ±1.92 pp | $10 | 1,340.05source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 87.542 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 10.000136 \| USD \| per_test \| Lower is better \| \| `latency` \| 1340.046 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.92 pp sourcePopulation 81 stderr 1.919 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-opus-4-7 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-opus-4-7` | | | | | | | | 10 | [GPT-6 Astra](https://benchize.com/models/gpt-6-astra-max.md)Max | 87.54% | SE ±1.92 pp | $23.92 | 1,512.6source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 87.542 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 23.924481 \| USD \| per_test \| Lower is better \| \| `latency` \| 1512.602 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.92 pp sourcePopulation 81 stderr 1.919 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-6-astra Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-6-astra` | | | | | | | | 12 | [Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max | 87.21% | SE ±1.94 pp | $16.63 | 1,555.02source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 87.205 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 16.626886 \| USD \| per_test \| Lower is better \| \| `latency` \| 1555.02 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.94 pp sourcePopulation 81 stderr 1.942 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-fable-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-fable-5` | | | | | | | | 13 | [Step 5 Preview](https://benchize.com/models/step-5-preview.md) | 86.53% | SE ±1.98 pp | $0.84 | 1,482.31source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 86.532 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.835096 \| USD \| per_test \| Lower is better \| \| `latency` \| 1482.306 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 1024000 provider Stepfun reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±1.98 pp sourcePopulation 81 stderr 1.984 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label stepfun/step-5-preview Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/stepfun~1step-5-preview` | | | | | | | | 14 | [DeepSeek V4 Flash 0731](https://benchize.com/models/deepseek-v4-flash-0731-high.md)High | 86.2% | SE ±2.01 pp | $0.05 | 654.63source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 86.195 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.053956 \| USD \| per_test \| Lower is better \| \| `latency` \| 654.63 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 384000 provider DeepSeek reasoning Not reported Reasoning effort high Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.01 pp sourcePopulation 81 stderr 2.005 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label deepseek/deepseek-v4-flash-0731 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/deepseek~1deepseek-v4-flash-0731` | | | | | | | | 14 | [DeepSeek V4.1 Flash](https://benchize.com/models/deepseek-v4-1-flash-max.md)Max | 86.2% | SE ±2.01 pp | $0.15 | 772.2source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 86.195 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.146247 \| USD \| per_test \| Lower is better \| \| `latency` \| 772.198 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 384000 provider DeepSeek reasoning Not reported Reasoning effort max Temperature 1 topP 1 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.01 pp sourcePopulation 81 stderr 2.005 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label deepseek/deepseek-v4.1-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/deepseek~1deepseek-v4.1-flash` | | | | | | | | 16 | [MiMo-V2.6-Flash](https://benchize.com/models/mimo-v2-6-flash.md) | 85.52% | SE ±2.05 pp | $0.07 | 758.7source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 85.522 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.07138 \| USD \| per_test \| Lower is better \| \| `latency` \| 758.704 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider Xiaomi reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.05 pp sourcePopulation 81 stderr 2.045 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label xiaomi/mimo-v2.6-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/xiaomi~1mimo-v2.6-flash` | | | | | | | | 17 | [GPT-6.1 Sol](https://benchize.com/models/gpt-6-1-sol-max.md)Max | 85.19% | SE ±2.07 pp | $2.3 | 1,588.49source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 85.185 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.297079 \| USD \| per_test \| Lower is better \| \| `latency` \| 1588.488 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.07 pp sourcePopulation 81 stderr 2.065 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-6.1-sol Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-6.1-sol` | | | | | | | | 18 | [Gemini 3.7 Flash](https://benchize.com/models/gemini-3-7-flash-high.md)High | 84.85% | SE ±2.08 pp | $2.38 | 683.69source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 84.848 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.379245 \| USD \| per_test \| Lower is better \| \| `latency` \| 683.694 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.08 pp sourcePopulation 81 stderr 2.084 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.7-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.7-flash` | | | | | | | | 19 | [Claude Sonnet 5](https://benchize.com/models/claude-sonnet-5-max.md)Max | 84.51% | SE ±2.1 pp | $6.18 | 2,812.68source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 84.512 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 6.176326 \| USD \| per_test \| Lower is better \| \| `latency` \| 2812.684 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.1 pp sourcePopulation 81 stderr 2.103 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-sonnet-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-sonnet-5` | | | | | | | | 20 | [Claude Sonnet 4.6](https://benchize.com/models/claude-sonnet-4-6-max.md)Max | 84.18% | SE ±2.12 pp | $2.78 | 1,127.79source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 84.175 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.783077 \| USD \| per_test \| Lower is better \| \| `latency` \| 1127.786 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning true Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.12 pp sourcePopulation 81 stderr 2.121 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-sonnet-4-6 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-sonnet-4-6` | | | | | | | | 21 | [Gemini 3.6 Flash](https://benchize.com/models/gemini-3-6-flash-high.md)High | 83.84% | SE ±2.14 pp | $2.14 | 906.2source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 83.838 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.137307 \| USD \| per_test \| Lower is better \| \| `latency` \| 906.196 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.14 pp sourcePopulation 81 stderr 2.14 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.6-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.6-flash` | | | | | | | | 21 | [Claude Opus 5](https://benchize.com/models/claude-opus-5-max.md)Max | 83.84% | SE ±2.14 pp | $16.15 | 2,613.81source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 83.838 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 16.147201 \| USD \| per_test \| Lower is better \| \| `latency` \| 2613.808 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.14 pp sourcePopulation 81 stderr 2.14 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-opus-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-opus-5` | | | | | | | | 23 | [DeepSeek V4 Pro](https://benchize.com/models/deepseek-v4-pro-max.md)Max | 83.5% | SE ±2.16 pp | $0.6 | 389.94source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 83.502 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.597321 \| USD \| per_test \| Lower is better \| \| `latency` \| 389.944 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 384000 provider DeepSeek reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.16 pp sourcePopulation 81 stderr 2.157 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label deepseek/deepseek-v4-pro Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/deepseek~1deepseek-v4-pro` | | | | | | | | 24 | [DeepSeek V4 Pro 0813](https://benchize.com/models/deepseek-v4-pro-0813-max.md)Max | 82.83% | SE ±2.19 pp | $0.15 | 1,069.22source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 82.828 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.148855 \| USD \| per_test \| Lower is better \| \| `latency` \| 1069.216 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 384000 provider DeepSeek reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.19 pp sourcePopulation 81 stderr 2.192 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label deepseek/deepseek-v4-pro-0813 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/deepseek~1deepseek-v4-pro-0813` | | | | | | | | 25 | [Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max | 82.49% | SE ±2.21 pp | $9.06 | 1,200.23source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 82.492 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 9.056848 \| USD \| per_test \| Lower is better \| \| `latency` \| 1200.23 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.21 pp sourcePopulation 81 stderr 2.209 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-opus-4-8 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-opus-4-8` | | | | | | | | 26 | [GPT-6 Luna](https://benchize.com/models/gpt-6-luna-max.md)Max | 82.16% | SE ±2.23 pp | $0.47 | 1,489.37source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 82.155 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.473577 \| USD \| per_test \| Lower is better \| \| `latency` \| 1489.37 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.23 pp sourcePopulation 81 stderr 2.226 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-6-luna Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-6-luna` | | | | | | | | 27 | [Claude Fable 5.1](https://benchize.com/models/claude-fable-5-1-max.md)Max | 80.14% | SE ±2.32 pp | $42.36 | 8,879.84source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 80.135 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 42.363937 \| USD \| per_test \| Lower is better \| \| `latency` \| 8879.844 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.32 pp sourcePopulation 81 stderr 2.319 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-fable-5-1 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-fable-5-1` | | | | | | | | 28 | [Claude Opus 5.5](https://benchize.com/models/claude-opus-5-5-max.md)Max | 78.79% | SE ±2.38 pp | $24.22 | 3,362.74source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 78.788 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 24.223778 \| USD \| per_test \| Lower is better \| \| `latency` \| 3362.74 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.38 pp sourcePopulation 81 stderr 2.376 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-opus-5-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-opus-5-5` | | | | | | | | 29 | [MiMo-V2.5-Pro](https://benchize.com/models/mimo-v2-5-pro.md) | 78.45% | SE ±2.39 pp | $0.1 | 603.84source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 78.451 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.099168 \| USD \| per_test \| Lower is better \| \| `latency` \| 603.842 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider Xiaomi reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.39 pp sourcePopulation 81 stderr 2.39 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label xiaomi/mimo-v2.5-pro Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/xiaomi~1mimo-v2.5-pro` | | | | | | | | 29 | [GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max | 78.45% | SE ±2.39 pp | $10.41 | 1,199.13source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 78.451 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 10.413192 \| USD \| per_test \| Lower is better \| \| `latency` \| 1199.134 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.39 pp sourcePopulation 81 stderr 2.39 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-5.6-sol Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-5.6-sol` | | | | | | | | 31 | [GPT-5.6 Luna](https://benchize.com/models/gpt-5-6-luna-max.md)Max | 78.11% | SE ±2.4 pp | $0.3 | 605.6source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 78.114 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.303629 \| USD \| per_test \| Lower is better \| \| `latency` \| 605.6 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.4 pp sourcePopulation 81 stderr 2.403 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-5.6-luna Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-5.6-luna` | | | | | | | | 32 | [GPT-6 Sol](https://benchize.com/models/gpt-6-sol-max.md)Max | 76.09% | SE ±2.48 pp | $2.68 | 732.28source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 76.094 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.683246 \| USD \| per_test \| Lower is better \| \| `latency` \| 732.282 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.48 pp sourcePopulation 81 stderr 2.479 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-6-sol Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-6-sol` | | | | | | | | 33 | [Gemini 3.5 Flash](https://benchize.com/models/gemini-3-5-flash-high.md)High | 73.74% | SE ±2.56 pp | $1.77 | 380.55source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 73.737 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 1.771686 \| USD \| per_test \| Lower is better \| \| `latency` \| 380.548 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.56 pp sourcePopulation 81 stderr 2.558 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.5-flash Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.5-flash` | | | | | | | | 34 | [GPT-5.6 Terra](https://benchize.com/models/gpt-5-6-terra-max.md)Max | 73.4% | SE ±2.57 pp | $2.56 | 1,094.4source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 73.401 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 2.560117 \| USD \| per_test \| Lower is better \| \| `latency` \| 1094.404 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.57 pp sourcePopulation 81 stderr 2.568 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-5.6-terra Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-5.6-terra` | | | | | | | | 35 | [MiniMax M2.7](https://benchize.com/models/minimax-m2-7.md) | 71.72% | SE ±2.62 pp | $0.59 | 445.29source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 71.717 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.588553 \| USD \| per_test \| Lower is better \| \| `latency` \| 445.292 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 80000 provider MiniMax reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.62 pp sourcePopulation 81 stderr 2.618 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label minimax/MiniMax-M2.7 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/minimax~1MiniMax-M2.7` | | | | | | | | 36 | [Claude Haiku 5.5](https://benchize.com/models/claude-haiku-5-5-max.md)Max | 67.68% | SE ±2.72 pp | $3.63 | 2,284.89source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 67.677 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 3.632901 \| USD \| per_test \| Lower is better \| \| `latency` \| 2284.886 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.72 pp sourcePopulation 81 stderr 2.719 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-haiku-5-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-haiku-5-5` | | | | | | | | 37 | [Nemotron 3 Ultra](https://benchize.com/models/nemotron-3-ultra-550b-a55b.md) | 67.34% | SE ±2.73 pp | — | 2,326source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 67.34 \| percent \| reported \| Higher is better \| \| `latency` \| 2326 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider Nvidia reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.73 pp sourcePopulation 81 stderr 2.726 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label nvidia/nemotron-3-ultra-550b-a55b Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/nvidia~1nemotron-3-ultra-550b-a55b` | | | | | | | | 38 | [GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high | 65.32% | SE ±2.77 pp | $3.62 | 571.98source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 65.32 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 3.615019 \| USD \| per_test \| Lower is better \| \| `latency` \| 571.98 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider OpenAI reasoning Not reported Reasoning effort xhigh Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.77 pp sourcePopulation 81 stderr 2.766 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label openai/gpt-5.5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/openai~1gpt-5.5` | | | | | | | | 39 | [Gemini 3.5 Flash-Lite](https://benchize.com/models/gemini-3-5-flash-lite-high.md)High | 64.65% | SE ±2.78 pp | $0.26 | 282.53source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 64.646 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.260621 \| USD \| per_test \| Lower is better \| \| `latency` \| 282.53 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.78 pp sourcePopulation 81 stderr 2.779 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.5-flash-lite Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.5-flash-lite` | | | | | | | | 39 | [Claude Sonnet 5.5](https://benchize.com/models/claude-sonnet-5-5-max.md)Max | 64.65% | SE ±2.78 pp | $17.21 | 3,657.38source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 64.646 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 17.206836 \| USD \| per_test \| Lower is better \| \| `latency` \| 3657.376 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort max harness Not reported maxOutputTokens 128000 provider Anthropic reasoning Not reported Reasoning effort max Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.78 pp sourcePopulation 81 stderr 2.779 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label anthropic/claude-sonnet-5-5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/anthropic~1claude-sonnet-5-5` | | | | | | | | 41 | [MiMo-V2.5](https://benchize.com/models/mimo-v2-5.md) | 63.97% | SE ±2.79 pp | $0.03 | 439.74source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 63.973 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.028794 \| USD \| per_test \| Lower is better \| \| `latency` \| 439.742 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 128000 provider Xiaomi reasoning Not reported Reasoning effort unknown Temperature 1 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.79 pp sourcePopulation 81 stderr 2.79 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label xiaomi/mimo-v2.5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/xiaomi~1mimo-v2.5` | | | | | | | | 42 | [Muse Spark Open](https://benchize.com/models/muse-spark-open-xhigh.md)Extra high | 56.23% | SE ±2.88 pp | $3.52 | 7,655.6source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 56.229 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 3.523605 \| USD \| per_test \| Lower is better \| \| `latency` \| 7655.598 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 131072 provider Together AI reasoning Not reported Reasoning effort xhigh Temperature 1 topP 1 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.88 pp sourcePopulation 81 stderr 2.884 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label meta/muse-spark-open Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/meta~1muse-spark-open` | | | | | | | | 42 | [Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview-high.md)High | 56.23% | SE ±2.88 pp | $0.67 | 275.23source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 56.229 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.665432 \| USD \| per_test \| Lower is better \| \| `latency` \| 275.234 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.88 pp sourcePopulation 81 stderr 2.884 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.1-pro-preview Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.1-pro-preview` | | | | | | | | 44 | [Mistral Medium 3.5](https://benchize.com/models/mistral-medium-3-5-high.md)High | 54.55% | SE ±2.89 pp | $5.19 | 1,835.93source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 54.545 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 5.191854 \| USD \| per_test \| Lower is better \| \| `latency` \| 1835.934 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 80000 provider Mistral AI reasoning Not reported Reasoning effort high Temperature 0.7 topP 0.95 verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.89 pp sourcePopulation 81 stderr 2.894 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label mistralai/mistral-medium-3.5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/mistralai~1mistral-medium-3.5` | | | | | | | | 45 | [Laguna M 1](https://benchize.com/models/laguna-m-1.md) | 44.44% | SE ±2.89 pp | $0.35 | 286.14source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 44.444 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.345186 \| USD \| per_test \| Lower is better \| \| `latency` \| 286.138 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 32768 provider Poolside reasoning Not reported Reasoning effort unknown Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.89 pp sourcePopulation 81 stderr 2.888 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label poolside/laguna-m.1 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/poolside~1laguna-m.1` | | | | | | | | 46 | [Mercury 2.5](https://benchize.com/models/mercury-2-5-high.md)High | 41.08% | SE ±2.86 pp | $0.18 | 119.38source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 41.077 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.176704 \| USD \| per_test \| Lower is better \| \| `latency` \| 119.378 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Inception reasoning Not reported Reasoning effort high Temperature Not reported topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.86 pp sourcePopulation 81 stderr 2.86 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label inception/mercury-2.5 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/inception~1mercury-2.5` | | | | | | | | 47 | [Gemini 3 Flash Preview](https://benchize.com/models/gemini-3-flash-preview-high.md)High | 39.06% | SE ±2.84 pp | $0.24 | 136.07source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 39.057 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.243093 \| USD \| per_test \| Lower is better \| \| `latency` \| 136.074 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.84 pp sourcePopulation 81 stderr 2.836 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3-flash-preview Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3-flash-preview` | | | | | | | | 48 | [Laguna Xs 2](https://benchize.com/models/laguna-xs-2.md) | 35.02% | SE ±2.77 pp | $0.14 | 440.76source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 35.017 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.136331 \| USD \| per_test \| Lower is better \| \| `latency` \| 440.762 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 32768 provider Poolside reasoning Not reported Reasoning effort unknown Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.77 pp sourcePopulation 81 stderr 2.773 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label poolside/laguna-xs.2 Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/poolside~1laguna-xs.2` | | | | | | | | 49 | [Gemini 3.1 Flash Lite Preview](https://benchize.com/models/gemini-3-1-flash-lite-preview-high.md)High | 27.95% | SE ±2.61 pp | $0.07 | 47.61source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 27.946 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.072372 \| USD \| per_test \| Lower is better \| \| `latency` \| 47.606 \| source_duration \| reported_duration \| Lower is better \| ### Configuration computeEffort Not reported harness Not reported maxOutputTokens 65536 provider Google reasoning Not reported Reasoning effort high Temperature 1 topP Not reported verbosity Not reported ### Reported statistics Uncertainty Standard error: ±2.61 pp sourcePopulation 81 stderr 2.608 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_legal_agent_benchmark_v1.ZWRzZKcv.json) Source Vals AI Original model label google/gemini-3.1-flash-lite-preview Source version 1 Observed 2026-10-21T16:45:53.909Z Evidence SHA-256 `65e7cd51e83066a42c9d782635dba5f2e9e1b189bd8118d3510cd17d2ddcba3f` Source locator `/tasks/criteria_pass_rate__white_collar_defense_investigations/google~1gemini-3.1-flash-lite-preview` | | | | | | | --- Cite as: Harvey's Legal Agent Benchmark v1 / criteria pass rate white collar defense investigations — Benchize, https://benchize.com/benchmarks/vals-ai%3Ahlab-v1%3Acriteria_pass_rate__white_collar_defense_investigations (retrieved 2026-10-21)