AI Model Capability Comparison
Science & math
Only models released within the past year are shown. Only benchmarks covering at least 3 of these models are shown. Unknown release dates are excluded; historical data is retained.
— means no published score for this candidate in this source, not zero.
| No. | Model | 10/11 recent-year models have scores | 10/11 recent-year models have scores | 10/11 recent-year models have scores | 10/11 recent-year models have scores | 9/11 recent-year models have scores | 4/11 recent-year models have scores |
|---|---|---|---|---|---|---|---|
| 1 |  Claude Opus 5.5 | 31.71maxclaude-opus-5-5_max | 91.23maxclaude-opus-5-5_max | 95maxclaude-opus-5-5_max | 100maxclaude-opus-5-5_max | 90.59maxclaude-opus-5-5_max | — |
| 2 |  GPT-6 Astra | 31.71maxgpt-6-astra_max | 93.68maxgpt-6-astra_max | 97.6highgpt-6-astra_high | 99unknowngpt-6-astra_unknown | 95.77maxgpt-6-astra_max | 54.8unknowngpt-6-astra_unknown |
| 3 | GPT-6.1 Sol | 31.71maxgpt-6.1-sol_max | 93.68maxgpt-6.1-sol_max | 100maxgpt-6.1-sol_max | 99unknowngpt-6.1-sol_unknown | 95.39maxgpt-6.1-sol_max | — |
| 4 | Claude Fable 5.1 | 31.14xhighclaude-fable-5-1_xhigh | 90.18maxclaude-fable-5-1_max | 87.8maxclaude-fable-5-1_max | 100maxclaude-fable-5-1_max | — | 46.5xhighclaude-fable-5-1_xhigh |
| 5 |  Gemini 4 Argon | 27.14highgemini-4-argon_high | 85.96highgemini-4-argon_high | 78.05highgemini-4-argon_high | 99unknowngemini-4-argon_unknown | 96.4highgemini-4-argon_high | — |
| 6 | Claude Sonnet 5.5 | 31.43maxclaude-sonnet-5-5_max | 88.77maxclaude-sonnet-5-5_max | 80.49maxclaude-sonnet-5-5_max | 100maxclaude-sonnet-5-5_max | 95.58maxclaude-sonnet-5-5_max | 46.12xhighclaude-sonnet-5-5_xhigh |
| 7 | Claude Opus 5 | 29.14maxclaude-opus-5_max | 85.61maxclaude-opus-5_max | 73.17maxclaude-opus-5_max | 99maxclaude-opus-5_max | 93.88maxclaude-opus-5_max | — |
| 8 | GPT-6 Sol | 30.86maxgpt-6-sol_max | 89.82maxgpt-6-sol_max | 90maxgpt-6-sol_max | 83unknowngpt-6-sol_unknown | 94.32maxgpt-6-sol_max | — |
| 9 | GPT-6 Luna | 19.43maxgpt-6-luna_max | 78.95maxgpt-6-luna_max | 56.1maxgpt-6-luna_max | 64unknowngpt-6-luna_unknown | 90.47maxgpt-6-luna_max | — |
| 10 | Claude Haiku 5.5 | 22.57xhighclaude-haiku-5-5_xhigh | 75.09maxclaude-haiku-5-5_max | 46.34maxclaude-haiku-5-5_max | 89maxclaude-haiku-5-5_max | 89.65maxclaude-haiku-5-5_max | 30.12unknownclaude-haiku-5-5_unknown |
| 11 | Claude Mythos Preview | — | — | — | — | — | — |
Epoch AI benchmarks
Source: Epoch AI, CC BY 4.0 ↗ · Updated 2026-10-15
CritPt · Updated 2026-10-15
FrontierMath Tiers 1-3 · Updated 2026-10-15
FrontierMath Tier 4 · Updated 2026-10-15
ProofBench · Updated 2026-10-12
GPQA Diamond · Updated 2026-10-15
Humanity's Last Exam · Updated 2026-10-07
PandaNpc Updated Oct 21, 2026, 06:01 UTC
Source: Epoch AI, CC BY 4.0 · Vals.ai · Artificial Analysis (artificialanalysis.ai)
Column guide
- No.
- Display order only. Sorting changes this number; it is not a published capability rank.
- Model
- The evaluated model name and its provider icon. A dash in a score column means no published score is available here, not a zero.
- CritPtResearch-level physics problems
- Tests reasoning on unpublished research-level physics problems across modern physics fields. Evaluation methodology ↗
- FrontierMath Tiers 1-3Frontier mathematics — tiers 1-3
- Tests challenging original mathematics problems across difficulty tiers 1–3. These scores should not be merged with the harder tier 4. Evaluation methodology ↗
- FrontierMath Tier 4Frontier mathematics — tier 4
- Tests the most difficult tier of FrontierMath research-level problems, separately from tiers 1–3. Evaluation methodology ↗
- ProofBenchFormal mathematical proof benchmark
- Tests producing mathematical proofs that can be checked by a formal verifier. Evaluation methodology ↗
- GPQA DiamondGraduate-level question answering — Diamond
- Tests difficult graduate-level science questions requiring specialist knowledge and reasoning. Evaluation methodology ↗
- Humanity's Last ExamHumanity final exam
- Tests difficult expert-level questions across many academic subjects. It is a question-answer benchmark, distinct from Agents' Last Exam. Evaluation methodology ↗
Read scores within the same benchmark, source and version. Higher values indicate stronger performance on the reported metric, but different columns do not share a common scale.