The rest of the field
Other models
The 6 pins that run every round but do not lead the home page. Same prompts, same hourly pulse, same units as the 4 models on the main boards.
The models we lead with
Fable 5, Opus 5, GPT-5.6 Sol, and Grok 4.6 lead the main boards on the home page.
Speed
Speed, ranked
These 6 models, ranked on the middle wall tok/s of their clean rounds in the last 24 hours. Longer bar is faster.
Tokens per second, start to finish. Higher is faster.
| Rank | Model | Bar, on one shared scale | Wall tok/s | Clean | Rounds |
|---|---|---|---|---|---|
| 1 | Haiku 4.5Anthropic · effort high · last round rate limit warning | 59.17 | 97% | 32/33 | |
| 2 | Grok 4.5xAI · effort high | 41.39 | 100% | 32/32 | |
| 3 | Sonnet 5Anthropic · effort high · last round rate limit warning | 32.06 | 97% | 32/33 | |
| 4 | GPT-5.5OpenAI · effort high | 24.66 | 100% | 33/33 | |
| 5 | GPT-5.6 TerraOpenAI · effort high | 22.54 | 94% | 31/33 | |
| 6 | GPT-5.6 LunaOpenAI · effort high | 22.53 | 100% | 33/33 |
Window 24 hours · 6 of 6 models scored · n = 197 rounds tried · test kit v6. Day-over-day change appears here once the lab has two full days of rounds in the log. 6 of 10 pinned models shown · the 4 we lead with are on the live board.
1 retired pin
Retired pins
Pins we can no longer run. Their pages stay up, and their old rounds stay in Logs.
GPT-5.2retired
Retired on Aug 19, 2026. OpenAI’s backend now refuses this model on subscription accounts, so our lab cannot test it anymore.