Toloka Arena (v1) pass^1
Agents
Per-trial success rate on Toloka Arena's superseded v1 task set, from the same five trials per task as its composite pass^5 headline: pass^1 is the share of single trials that succeed, the mean over the five trials.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 69.8 |
| Claude Fable 5.1 | 68.2 |
| Gemini 3.7 Flash | 66.9 |
| Gemini 3.6 Flash | 66.1 |
| Opus 4.8 | 65.7 |
| GPT-5.5 | 64.6 |
| GPT-6 Astra | 64.5 |
| Claude Opus 5 | 64.4 |
| Grok 4.5 | 64.1 |
| GPT-5.4 | 63.6 |
| Grok 4.6 | 63.5 |
| Opus 4.7 | 62.1 |
| GPT-5.6 Sol | 60.2 |
| GLM-5.2 | 59.2 |
| GLM-5.3 | 58.7 |
| GPT-5.6 Terra | 57.7 |
| Sonnet 5 | 57.5 |
| Grok 4.7 | 57.3 |
| Opus 4.6 | 56.4 |
| Kimi K3 | 54.4 |
| Gemini 3.5 Flash | 52.9 |
| Muse Spark 1.1 | 52.8 |
| Muse Spark 1.2 | 51.5 |
| Gemini 3.1 Pro Preview | 51.4 |
| Sonnet 4.6 | 50.2 |
| GLM-5.1 | 48.5 |
| Gemini 3 Flash Preview | 46.7 |
| Kimi K2.6 | 37.4 |
| MiniMax M3 | 34.0 |
| Mistral Medium 3.5 | 32.2 |
| gpt-oss-120b | 10.6 |
| Grok 4.20 | 6.7 |
| Nova 2 Lite | 3.7 |