# GPT-5.2Extra high > Source: Benchize (https://benchize.com) — https://benchize.com/models/gpt-5-2-xhigh. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/models/gpt-5-2-xhigh. OpenAI Canonical URL: https://benchize.com/models/gpt-5-2-xhigh 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. OpenAI ## Benchmarks | Benchmark | Score | ± | Rank | Cost | Source | | | --- | --- | --- | --- | --- | --- | --- | | [HMMT Nov 2025](https://benchize.com/benchmarks/matharena%3Ahmmt--hmmt_nov_2025.md) | 99.17% | 95% CI 97.54–100.8% | 1 / 12 | 0.19usd per problem | [matharena · hmmt--hmmt_nov_2025](https://matharena.ai/competition_tables/hmmt--hmmt_nov_2025) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 99.17 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 0.19 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition hmmt--hmmt_nov_2025 predicted false Reasoning effort xhigh sourceModelName GPT-5.2 (xhigh) sourceModelVersion openai/gpt-52 thinkingEnabled true ### Reported statistics Uncertainty 95% confidence interval: 97.54–100.8% confidenceInterval95 1.63 problems 30 ### Source [Original source](https://matharena.ai/competition_tables/hmmt--hmmt_nov_2025) Source matharena Original model label GPT-5.2 (xhigh) Source version hmmt--hmmt_nov_2025 Observed 2026-10-21T16:52:37.292Z Evidence SHA-256 `aa9287d733e27dde3646179a18070c07f558fb51926a8fe90cdf38985d6fce6d` Source locator `/table/rows/0` | | | | | | | | [SMT 2025](https://benchize.com/benchmarks/matharena%3Asmt--smt_2025.md) | 96.03% | — | 1 / 25 | 0.16usd per problem | [matharena · smt--smt_2025](https://matharena.ai/competition_tables/smt--smt_2025) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 96.03 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 0.16 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition smt--smt_2025 predicted true Reasoning effort xhigh sourceModelName GPT-5.2 (xhigh) sourceModelVersion openai/gpt-52 thinkingEnabled true ### Reported statistics confidenceInterval95 Not reported problems 53 ### Source [Original source](https://matharena.ai/competition_tables/smt--smt_2025) Source matharena Original model label GPT-5.2 (xhigh) Source version smt--smt_2025 Observed 2026-10-21T16:52:36.547Z Evidence SHA-256 `1b4f6f8bafc1688e4557a024513f68befa31aff48fd0623fb4892a61263433b0` Source locator `/table/rows/0` | | | | | | | | [MCP Atlas](https://benchize.com/benchmarks/scale%3Amcp_atlas.md) | 67.6source score | CI 64.7–70.5 | 15 / 22 | — | [scale](https://labs.scale.com/leaderboard/mcp_atlas) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 67.6 \| source_score \| reported \| Higher is better \| ### Configuration Reasoning effort xhigh Model configuration gpt-5.2 (xhigh) sourceModelVersion Not reported thinkingEnabled true ### Reported statistics Uncertainty Confidence interval: 64.7–70.5 contaminationWarning Not reported rank 24 reportedConfidenceHalfWidth 2.9 ### Source [Original source](https://labs.scale.com/leaderboard/mcp_atlas) Source scale Original model label gpt-5.2 (xhigh) Observed 2026-10-21T16:35:30.627Z Evidence SHA-256 `1467b1d078bf39a76fb663f006002e7c7d08224ca7c3de883ab628a2efccebb8` Source locator `/rows/23` | | | | | | | | [AIME 2025](https://benchize.com/benchmarks/aime-2025.md) | 100% | 95% CI 100–100% | 1 / 35 | 0.18usd per problem | [matharena · aime--aime_2025](https://matharena.ai/competition_tables/aime--aime_2025) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 100.00 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 0.18 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition aime--aime_2025 predicted false Reasoning effort xhigh sourceModelName GPT-5.2 (xhigh) sourceModelVersion openai/gpt-52 thinkingEnabled true ### Reported statistics Uncertainty 95% confidence interval: 100–100% confidenceInterval95 0.00 problems 30 ### Source [Original source](https://matharena.ai/competition_tables/aime--aime_2025) Source matharena Original model label GPT-5.2 (xhigh) Source version aime--aime_2025 Observed 2026-10-21T16:52:35.082Z Evidence SHA-256 `fab0abcecc4032852ef2cbd2a84d34ffc42202be358b2994bfaba44ae7a0cd9a` Source locator `/table/rows/1` | | | | | | | | [ARC-AGI-1](https://benchize.com/benchmarks/arc-prize%3Av1_Semi_Private.md) | 86.17% | — | 80 / 213 | 0.96usd per task | [arc prize · v1](https://arcprize.org/media/data/leaderboard/v1.json) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 0.8617 \| score_fraction \| v1_Semi_Private \| Higher is better \| \| `cost_per_task` \| 0.96 \| usd_per_task \| v1_Semi_Private \| Lower is better \| ### Configuration evaluationSplit v1_Semi_Private harness Not reported modelGroup gpt-5-2-2025-12-11-thinking modelType CoT Reasoning effort xhigh sourceModelVersion gpt-5-2-2025-12-11-thinking-xhigh thinkingEnabled true ### Source [Original source](https://arcprize.org/media/data/leaderboard/v1.json) Source arc prize Original model label GPT-5.2 (XHigh) Source version v1 Observed 2026-10-21T16:34:20.378Z Evidence SHA-256 `302d8267c0522bf450ec1deeac4d060e2cf2fb444ba0a4269014d9535de0bd1f` Source locator `/evaluations/103` | | | | | | | | [ARC-AGI-2](https://benchize.com/benchmarks/arc-prize%3Av2_Semi_Private.md) | 52.91% | — | 81 / 212 | 1.9usd per task | [arc prize · v2](https://arcprize.org/media/data/leaderboard/v2.json) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 0.5291 \| score_fraction \| v2_Semi_Private \| Higher is better \| \| `cost_per_task` \| 1.8982 \| usd_per_task \| v2_Semi_Private \| Lower is better \| ### Configuration evaluationSplit v2_Semi_Private harness Not reported modelGroup gpt-5-2-2025-12-11-thinking modelType CoT Reasoning effort xhigh sourceModelVersion gpt-5-2-2025-12-11-thinking-xhigh thinkingEnabled true ### Source [Original source](https://arcprize.org/media/data/leaderboard/v2.json) Source arc prize Original model label GPT-5.2 (XHigh) Source version v2 Observed 2026-10-21T16:34:20.750Z Evidence SHA-256 `bca68e12cc9a01954ae21b641844c1ecbabe146cb24b87f0f8e9b1db27433c29` Source locator `/evaluations/107` | | | | | | | | [BrokenArXiv 02/2026](https://benchize.com/benchmarks/matharena%3Aarxiv_false--february.md) | 25.81% | 95% CI 18.11–33.51% | 4 / 15 | 0.61usd per problem | [matharena · arxiv_false--february](https://matharena.ai/competition_tables/arxiv_false--february) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 25.81 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 0.61 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition arxiv_false--february predicted false Reasoning effort xhigh sourceModelName GPT-5.2 (xhigh) sourceModelVersion openai/gpt-52 thinkingEnabled true ### Reported statistics Uncertainty 95% confidence interval: 18.11–33.51% confidenceInterval95 7.70 problems 31 ### Source [Original source](https://matharena.ai/competition_tables/arxiv_false--february) Source matharena Original model label GPT-5.2 (xhigh) Source version arxiv_false--february Observed 2026-10-21T16:52:23.532Z Evidence SHA-256 `a6bc6fc18da7dd181ed727da924f9569d3e323c31c1aba9a4d9fa3bff492bee3` Source locator `/table/rows/3` | | | | | | | | [WeirdML v3](https://benchize.com/benchmarks/weirdml%3Av3.md) | 0.05source score | Range 0.04–0.06 | 23 / 25 | — | [weirdml · b0120fe](https://htihle.github.io/assets/data/weirdml_v3.json) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 0.0500787722943723 \| source_score \| reported \| Higher is better \| \| `mean_final_best` \| 0.09720329675324675 \| source_score \| reported \| Higher is better \| ### Configuration budget {"kind":"equivalent_tokens","limit":50000000,"start":500000} harness codex_cli Reasoning effort xhigh sourceConfiguration gpt-5.2-xhigh-openrouter thinkingEnabled true ### Reported statistics Uncertainty Reported interval: 0.04–0.06 interval [0.040373215586946704,0.059638243704060394] intervalMethod {"assumptions":"Independent fresh runs; approximately normal scores; shared variance prior. Pool fitted within analysis mode, including incomplete models; pool estimation uncertainty is not propagated.","confidence":0.95,"draws":10000,"fixed_tasks":true,"method":"partially_pooled_variance","pool":{"degrees_of_freedom":1533,"fallback":false,"prior_df":4,"variance":0.0026383113392869583},"seed":20260915,"unit":"run"} runCount 86 taskCount 11 ### Source [Original source](https://htihle.github.io/assets/data/weirdml_v3.json) Source weirdml Original model label GPT-5.2 (xhigh) Source version b0120fe Observed 2026-10-21T16:33:04.394Z Evidence SHA-256 `82c3a8de423b23ca0af9761429299a2f21edb0b819e947fdc863f8b5ea4a49b8` Source locator `/rows/22` | | | | | | | | [DesignArena Website generation](https://benchize.com/benchmarks/designarena%3Awebsite.md) | 1,195elo | — | 55 / 123 | — | [designarena · public](https://www.designarena.ai/api/leaderboard#board=website) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `elo` \| 1195 \| elo \| website \| Higher is better \| ### Configuration arenaType models category website inputModality text Reasoning effort xhigh sourceModelVersion candycane thinkingEnabled true variationName public ### Reported statistics averageGenerationTimeMs 239415 battles 846 losses 386 standardError Not reported winRate 54.4 wins 460 ### Source [Original source](https://www.designarena.ai/api/leaderboard#board=website) Source designarena Original model label GPT-5.2 (xhigh) Source version public Observed 2026-10-21T16:37:57.757Z Evidence SHA-256 `51873e97776a3c5ae706b24cc2c60384793dd18398df970e213a8e7c276d8ac0` Source locator `/data/87` | | | | | | | | [HMMT Feb 2025](https://benchize.com/benchmarks/matharena%3Ahmmt--hmmt_feb_2025.md) | 100% | 95% CI 100–100% | 1 / 34 | 0.24usd per problem | [matharena · hmmt--hmmt_feb_2025](https://matharena.ai/competition_tables/hmmt--hmmt_feb_2025) | | | ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 100.00 \| percent \| competition \| Higher is better \| \| `cost_per_problem` \| 0.24 \| usd_per_problem \| competition \| Lower is better \| ### Configuration competition hmmt--hmmt_feb_2025 predicted false Reasoning effort xhigh sourceModelName GPT-5.2 (xhigh) sourceModelVersion openai/gpt-52 thinkingEnabled true ### Reported statistics Uncertainty 95% confidence interval: 100–100% confidenceInterval95 0.00 problems 30 ### Source [Original source](https://matharena.ai/competition_tables/hmmt--hmmt_feb_2025) Source matharena Original model label GPT-5.2 (xhigh) Source version hmmt--hmmt_feb_2025 Observed 2026-10-21T16:52:35.608Z Evidence SHA-256 `8554225c26eb7a3c2649943ee7cf6b404ec1f7441b20fbdcfccbf35b9a864864` Source locator `/table/rows/0` | | | | | | | --- Cite as: GPT-5.2Extra high — Benchize, https://benchize.com/models/gpt-5-2-xhigh (retrieved 2026-10-21)