Every frontier model, measured.
GPT-6.2 leads at 202 (GPT-4 = 100); the best open model, Kimi K3.1 Pro, is about 5 months behind the closed frontier. Capability of every major AI model since ChatGPT, from trusted third parties such as Epoch AI, METR and ARC Prize: closed and open-weight models, side by side.
Each dot is one model, scored from third-party test results. GPT-4 at launch = 100. Hover for details, click for the full profile.
Reigns at #1how long each model held the top score
Closed models
- GPT-6.21 so far
- Claude Fable 5.28
- Gemini 4 Carbon12
- Claude Opus 5.536
- GPT-6 Astra18
- Claude Fable 5.13
- Claude Opus 539
- Claude Fable 545
- GPT-5.547
- Claude Opus 4.77
- GPT-5.442
- Claude Opus 4.628
Show all reigns
Open-weight models
- Kimi K3.1 Pro26 so far
- DeepSeek V4.1 Flash 10176
- GLM-5.353
- Kimi K329
- GLM-5.241
- Kimi K2.657
- GLM-5.113
- Kimi K2.570
Show all reigns
Reigns use today’s Index scores: a model tested weeks after release counts as #1 from its release date (open models from when their weights were published), and on a live leaderboard the order on a given day could differ. A change is clear when the gap is more than twice the combined standard error.
Top models todaycurrent versions · by index score
All models →| Model | Weights | Index score | Released | |
|---|---|---|---|---|
| 1 | GPT-6.2 | Closed | Nov 2026 | |
| 2 | GPT-6.2 Astra | Closed | Nov 2026 | |
| 3 | Claude Fable 5.2 | Closed | Nov 2026 | |
| 4 | Closed | Oct 2026 | ||
| 5 | Claude Opus 5.5 | Closed | Sep 2026 | |
| 6 | Closed | Sep 2026 | ||
| 7 | GPT-6.1 Sol | Closed | Sep 2026 | |
| 8 | Closed | Nov 2026 | ||
| 9 | Claude Sonnet 5.5 | Closed | Sep 2026 | |
| 10 | GPT-6.1 Terra | Closed | Oct 2026 |
Third-party test results only. Versions a lab has replaced are left out.