Skip to main content
AI Choice Engine
← Benchmark sources

artificialanalysis.ai · benchmark source

Artificial Analysis

Independent evaluations across composite intelligence, hard reasoning, science reasoning, and agentic terminal tasks. Multiple boards live under one publisher.

Open artificialanalysis.ai5 boards on this publisher

Expert reasoning

GPQA Diamond

A difficult question set intended to test graduate-level scientific reasoning.

Why it matters

It helps separate broad reasoning ability from surface fluency on demanding knowledge questions.

Use it when

Your work includes difficult science reasoning and you can tolerate a narrow benchmark lens.

The limitation

Saturated: frontier models score ~93–96% and Artificial Analysis dropped it from its Intelligence Index in v4.3, so small gaps are noise. Scientific question answering is not the same as coding, writing, retrieval, or safe production use.

Leaderboard snapshot

GPQA Diamond results in a readable view

Use the tabs for the primary score and other published fields from this board. The source link remains authoritative for the live table.

Only three exact public AA headline configurations are retained. GPQA Diamond is near its ceiling and was dropped from Intelligence Index v4.3; the full component data require publisher access.
3 exact matching primary-source observations are ranked in this retained subset. 610 unsupported observations are excluded from ranks, charts and leader claims. Subset ranks are not positions on the complete publisher leaderboard.

Leader among verified rows

GPT-6 Astra (xhigh)

96.3%

Rows shown

3

Accuracy

Source / snapshot date

2026-10-01

25 days old · not a live API feed

Score profile

Accuracy by published row

Higher is better in this view

GPT-6 Astra (xhigh)
96.3%
GPT-6 Astra (max)
96.1%
Gemini 3.8 Flash (high)
95.3%
Not yet verified (610)

No exact matching primary-source observation was retained for these entries. Scores are withheld here; these entries do not affect ranks, leaders, charts or search indexing.

  • GPT-6 Astra (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.6 (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.7 Flash (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.1 Pro Preview

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Sol (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • Muse Spark 1.3 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-6 Astra (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 5 (Adaptive Reasoning, High Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.8 Flash (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.5 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.6 (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.6 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Kimi K3 (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • Muse Spark 1.3 (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.8 2.4T A95B

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 5 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.5 (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Sol (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-6 Astra (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.5 (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • MiniMax-M3

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Pro 0813 (Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.6 Flash (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Sol (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.8 Max (0902)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.8 Max

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.5 (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Sol (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Terra (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.7 Max

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.8-Flash-Next

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.5 Flash (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.5 Flash (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.7 Flash (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 4.8 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.8 Flash (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.4 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 5 (Adaptive Reasoning, Medium Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GLM-5.3 (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.3 Codex (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 4.7 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Flash Vision (Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GLM 5.3 Flash

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Sonnet 5 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Luna (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.20 0309 v2 (Reasoning)

    No exact matching primary-source row was retained; excluded from rankings.

  • Kimi K2.6

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.5 (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Flash 0731 (Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3 Pro Preview (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Terra (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Pro 0424 (Reasoning, High Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.8 27B (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Muse Spark 1.2 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.2 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3.7 Flash (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.3 (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.7 Plus

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.2 Codex (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3 Flash Preview (Reasoning)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Sol (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • Muse Spark 1.1 (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Hy3

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 4.6 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Terra (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Kimi K2.7 Code

    No exact matching primary-source row was retained; excluded from rankings.

  • GLM-5.2 (max)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Luna (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok Build 0.1 0616

    No exact matching primary-source row was retained; excluded from rankings.

  • Inkling Small

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Flash 0420 (Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.5 397B A17B (Reasoning)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Luna (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • Nex-N2-Pro (based on Qwen3.5-397B-A17B)

    No exact matching primary-source row was retained; excluded from rankings.

  • Solar Pro 4

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.3 (medium)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 5 (Adaptive Reasoning, Low Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • DeepSeek V4 Pro 0424 (Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.6 Max Preview

    No exact matching primary-source row was retained; excluded from rankings.

  • Gemini 3 Pro Preview (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5.1 (Adaptive Reasoning, Medium Effort, Default Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Opus 4.7 (Non-reasoning, High Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.20 0309 (Reasoning)

    No exact matching primary-source row was retained; excluded from rankings.

  • Muse Spark

    No exact matching primary-source row was retained; excluded from rankings.

  • Qwen3.6 Plus

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Fable 5.1 (Adaptive Reasoning, Low Effort, Default Fallback)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4.6 (low)

    No exact matching primary-source row was retained; excluded from rankings.

  • Kimi K2.5 (Reasoning)

    No exact matching primary-source row was retained; excluded from rankings.

  • Grok 4

    No exact matching primary-source row was retained; excluded from rankings.

  • Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.4 mini (xhigh)

    No exact matching primary-source row was retained; excluded from rankings.

  • MiniMax-M2.7

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.1 (high)

    No exact matching primary-source row was retained; excluded from rankings.

  • K2 Horizon 375B A23B

    No exact matching primary-source row was retained; excluded from rankings.

  • GPT-5.6 Terra (medium)

    No exact matching primary-source row was retained; excluded from rankings.

Showing 100 entries. Open the publisher source for the current complete table.

Only three exact public AA headline configurations are retained. GPQA Diamond is near its ceiling and was dropped from Intelligence Index v4.3; the full component data require publisher access. 3 exact matching primary-source observations are ranked in this retained subset. 610 unsupported observations are excluded from ranks, charts and leader claims. Subset ranks are not positions on the complete publisher leaderboard. Exact retained receipts captured 2026-10-01; source/run dates are shown per verified row when known.

Open Artificial Analysis GPQA Diamond leaderboard

Evidence in this catalog

Where GPQA Diamond fits

Closest verified examples we currently carry. A missing score is not a zero.

Model examples

  • GPT-6 Astra (xhigh)96.3%

    Published source observation captured 2026-10-01. Source-specific result; no independent rerun or universal capability verdict.

  • GPT-6 Astra (max)96.1%

    Published source observation captured 2026-10-01. Source-specific result; no independent rerun or universal capability verdict.

  • Gemini 3.8 Flash (high)95.3%

    Published source observation captured 2026-10-01. Source-specific result; no independent rerun or universal capability verdict.

Harness examples

No directly comparable harness score is verified here yet.

Open this board on Artificial Analysis GPQA Diamond leaderboard