Big claims.
Measured results.
See how AI models perform on the tests behind the headlines. Explore the scores, compare like for like, and follow the evidence.
Your next model, on your terms.
Upgrade Watch compares new results against the model and benchmarks you choose.
Know what deserves your next trial.
Save your current model, set score targets and see improvements, regressions and gaps side by side. Changes stay highlighted until you review them.
Saved across devices Original evaluation sourcesWhat counts as a match?
Every selected test must meet its gain target, with at least one improvement. We only flag targets when the recorded test version, named effort setting and report match. Missing settings, different reports and unspecified agent configurations are shown as Context needed, not upgrade signals. Matching effort labels do not establish equal compute.
These are adjusted benchmark scores, not a prediction of your workload. Epoch retains selected high scores rather than every test run; a missing result means it is not in this snapshot. No overall score is invented. How the source data is prepared ↗
Choose a benchmark.
Retrieved 2026-11-03 23:47 UTC
GPQA diamond
185 matching results · higher is betterAdjusted scores on a 0–100 scale, not raw accuracy percentages. Epoch rescales some tests for chance performance or known errors. Compare within the same benchmark. Model release dates are not test dates.
“Reported in” identifies the report containing the result; it can compare models from other companies. Effort is shown only when specified in the dataset.
| Model | Adjusted score | Model released | Reported in |
|---|---|---|---|
| GPT-6.1 TerraEffort: Max | 96.3/100 | Epoch AI datasetOriginal report not specified | |
| DeepSeek V4.1 Flash 1017Effort: Max | 94.7/100 | Epoch AI datasetOriginal report not specified | |
| Muse Spark OpenEffort: Extra high | 88.3/100 | Epoch AI datasetOriginal report not specified | |
| Claude Haiku 5.5Effort: Max | 89.8/100 | Epoch AI datasetOriginal report not specified | |
| Mistral Large 4Effort: High | 87.6/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 4 ArgonEffort: High | 99.2/100 | Epoch AI datasetOriginal report not specified | |
| GPT-6.1 SolEffort: Max | 97.8/100 | Epoch AI datasetOriginal report not specified | |
| Claude Sonnet 5.5Effort: Max | 98.0/100 | Epoch AI datasetOriginal report not specified | |
| GPT-6 SolEffort: Max | 96.3/100 | Epoch AI datasetOriginal report not specified | |
| Claude Opus 5.5Effort: Max | 91.1/100 | Epoch AI datasetOriginal report not specified | |
| GPT-6 LunaEffort: Max | 90.9/100 | Epoch AI datasetOriginal report not specified | |
| Grok 4.7Effort: Extra high | 94.1/100 | Epoch AI datasetOriginal report not specified | |
| DeepSeek V4.1 FlashEffort: Max | 90.0/100 | Epoch AI datasetOriginal report not specified | |
| GPT-6 AstraEffort: Max | 98.3/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 3.8 FlashEffort: High | 97.8/100 | Epoch AI datasetOriginal report not specified | |
| Qwen3.8 Max (0902)Effort: Extra high | 93.5/100 | Epoch AI datasetOriginal report not specified | |
| GLM-5.3-FlashEffort: Max | 90.5/100 | Epoch AI datasetOriginal report not specified | |
| GLM-5.3Effort: Max | 91.5/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 3.7 FlashEffort: High | 97.0/100 | Epoch AI datasetOriginal report not specified | |
| DeepSeek V4 Pro 0813Effort: Max | 92.6/100 | Epoch AI datasetOriginal report not specified | |
| Grok 4.6Effort: Extra high | 95.8/100 | Epoch AI datasetOriginal report not specified | |
| Qwen 3.8 MaxEffort: Extra high | 94.0/100 | Epoch AI datasetOriginal report not specified | |
| DeepSeek V4 Flash 0731Effort: Max | 91.7/100 | Epoch AI datasetOriginal report not specified | |
| Qwen3.7 FlashEffort unknown | 79.6/100 | Epoch AI datasetOriginal report not specified | |
| Claude Opus 5Effort unknown | 95.7/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 3.6 FlashEffort: Minimal | 96.0/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 3.5 Flash-LiteEffort: Low | 81.0/100 | Epoch AI datasetOriginal report not specified | |
| Kimi K3Effort: Low | 94.6/100 | Epoch AI datasetOriginal report not specified | |
| Inkling-SmallEffort: Extra high | 88.2/100 | Epoch AI datasetOriginal report not specified | |
| InklingEffort: Extra high | 87.9/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.6 SolEffort: None | 95.1/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.6 TerraEffort: None | 94.9/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.6 LunaEffort: None | 92.5/100 | Epoch AI datasetOriginal report not specified | |
| Grok 4.5Effort: High | 95.0/100 | Epoch AI datasetOriginal report not specified | |
| Claude Sonnet 5Effort: Max | 91.0/100 | Epoch AI datasetOriginal report not specified | |
| GLM-5.2Effort: None | 92.9/100 | Epoch AI datasetOriginal report not specified | |
| Kimi K2.7 CodeEffort unknown | 87.3/100 | Epoch AI datasetOriginal report not specified | |
| Claude Fable 5Effort: Max | 84.5/100 | Epoch AI datasetOriginal report not specified | |
| Nemotron 3 UltraEffort unknown | 83.8/100 | Epoch AI datasetOriginal report not specified | |
| Qwen3.7-PlusEffort unknown | 87.3/100 | Epoch AI datasetOriginal report not specified | |
| MiniMax-M3Effort: None | 91.5/100 | Epoch AI datasetOriginal report not specified | |
| Claude Opus 4.8Effort: None | 91.7/100 | Epoch AI datasetOriginal report not specified | |
| Gemini 3.5 FlashEffort: Low | 94.2/100 | Epoch AI datasetOriginal report not specified | |
| Qwen3.7-MaxEffort unknown | 91.5/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.5 InstantEffort unknown | 79.9/100 | Epoch AI datasetOriginal report not specified | |
| Qwen 3.6 FlashEffort unknown | 81.0/100 | Epoch AI datasetOriginal report not specified | |
| DeepSeek-V4-ProEffort: None | 91.5/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.5Effort: None | 95.8/100 | Epoch AI datasetOriginal report not specified | |
| GPT-5.5 ProEffort: Extra high | 95.7/100 | Epoch AI datasetOriginal report not specified | |
| Qwen3.6 27BEffort unknown | 84.5/100 | Epoch AI datasetOriginal report not specified |