# Public Benefits Bench v1 / multi turn > Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/vals-ai%3Apublic-benefits-bench-v1%3Amulti_turn. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/vals-ai%3Apublic-benefits-bench-v1%3Amulti_turn. Can AI help people navigate SNAP benefits? Canonical URL: https://benchize.com/benchmarks/vals-ai%3Apublic-benefits-bench-v1%3Amulti_turn 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. Can AI help people navigate SNAP benefits? Professional work · Models tested: 8 · Score publishers: Vals AI ## Model results | # | Model | Score | ± | Cost | Latency | | | --- | --- | --- | --- | --- | --- | --- | | 1 | [Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max | 65.22% | SE ±1.24 pp | $0.29 | 272.42source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 65.223 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.294256 \| USD \| per_test \| Lower is better \| \| `latency` \| 272.419 \| 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: ±1.24 pp sourcePopulation 13 stderr 1.239 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label anthropic/claude-fable-5 Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/anthropic~1claude-fable-5` | | | | | | | | 2 | [Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max | 58.46% | SE ±1.28 pp | $0.59 | 311.36source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 58.457 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.592054 \| USD \| per_test \| Lower is better \| \| `latency` \| 311.358 \| 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: ±1.28 pp sourcePopulation 13 stderr 1.282 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label anthropic/claude-opus-4-8 Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/anthropic~1claude-opus-4-8` | | | | | | | | 3 | [Gemini 3.5 Flash](https://benchize.com/models/gemini-3-5-flash-high.md)High | 53.99% | SE ±1.3 pp | $0.13 | 129.71source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 53.992 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.130219 \| USD \| per_test \| Lower is better \| \| `latency` \| 129.707 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.296 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label google/gemini-3.5-flash Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/google~1gemini-3.5-flash` | | | | | | | | 4 | [GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high | 53.86% | SE ±1.3 pp | $0.57 | 434.05source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 53.857 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.567811 \| USD \| per_test \| Lower is better \| \| `latency` \| 434.054 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.297 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label openai/gpt-5.5 Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/openai~1gpt-5.5` | | | | | | | | 5 | [Gemini 3.1 Pro Preview](https://benchize.com/models/gemini-3-1-pro-preview-high.md)High | 52.91% | SE ±1.3 pp | $0.09 | 109.31source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 52.909 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.087001 \| USD \| per_test \| Lower is better \| \| `latency` \| 109.306 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.298 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label google/gemini-3.1-pro-preview Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/google~1gemini-3.1-pro-preview` | | | | | | | | 6 | [Claude Sonnet 4.6](https://benchize.com/models/claude-sonnet-4-6-max.md)Max | 52.37% | SE ±1.3 pp | $0.28 | 363.46source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 52.368 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.278287 \| USD \| per_test \| Lower is better \| \| `latency` \| 363.464 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.299 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label anthropic/claude-sonnet-4-6 Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/anthropic~1claude-sonnet-4-6` | | | | | | | | 7 | [DeepSeek V4 Pro](https://benchize.com/models/deepseek-v4-pro-max.md)Max | 48.71% | SE ±1.3 pp | $0.09 | 350.63source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 48.714 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.086444 \| USD \| per_test \| Lower is better \| \| `latency` \| 350.63 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.3 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label deepseek/deepseek-v4-pro Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/deepseek~1deepseek-v4-pro` | | | | | | | | 8 | [MiniMax M3](https://benchize.com/models/minimax-m3.md) | 48.24% | SE ±1.3 pp | $0.09 | 492.82source duration | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 48.241 \| percent \| reported \| Higher is better \| \| `cost_per_test` \| 0.086954 \| USD \| per_test \| Lower is better \| \| `latency` \| 492.82 \| 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: ±1.3 pp sourcePopulation 13 stderr 1.3 tokenTotals Not reported ### Source [Original source](https://www.vals.ai/_astro/benchmark_view_snap_v1.BPa0Hrre.json) Source Vals AI Original model label minimax/MiniMax-M3 Source version 1 Observed 2026-10-21T16:46:18.363Z Evidence SHA-256 `b85e2b3cbdf03e13e2ffdb27ce4f108003a548d705a24ae26a0a115e2027475d` Source locator `/tasks/multi_turn/minimax~1MiniMax-M3` | | | | | | | --- Cite as: Public Benefits Bench v1 / multi turn — Benchize, https://benchize.com/benchmarks/vals-ai%3Apublic-benefits-bench-v1%3Amulti_turn (retrieved 2026-10-21)