A-Fantasia Chess (Valid Attempts with Retries)
Games · 2026-09-30
A-Fantasia chess error among valid responses under the retry protocol. A correct answer is any legal SAN move. Latest run with at least 80 valid items; up to five retries for unscorable responses; denominator is not published. The current scorer accepts an initial SAN token before commentary.
Top models (lower is better)
| Model | Score |
|---|---|
| GPT-4.5 | 3.0 |
| GPT-5 Chat | 11.0 |
| Gemini 2.0 Flash | 12.0 |
| GPT-4o (source unspecified) | 13.0 |
| GPT-4.1 | 13.0 |
| GPT-5.5 | 14.0 |
| Qwen3.7-Max (source unspecified) | 16.0 |
| GPT-5.1 | 16.0 |
| Opus 4.1 | 19.0 |
| Opus 4.6 | 20.0 |
| Opus 4.5 | 22.0 |
| GPT-5.6 Sol | 22.0 |
| GPT-5.6 Terra | 22.0 |
| Gemini 2.0 Flash-Lite | 22.0 |
| Sonnet 4.5 | 25.0 |
| GPT-5.6 Luna | 25.0 |
| Claude Opus 5 | 27.0 |
| Gemini 2.5 Flash | 27.0 |
| Grok 3 (API) | 28.0 |
| Opus 4 | 29.0 |
| Sonnet 4 | 29.0 |
| Hy4 Preview | 29.0 |
| Opus 4.8 | 30.0 |
| Sonnet 4.6 | 30.0 |
| Kimi K2.6 | 32.0 |
| GPT-5.4 | 33.0 |
| Claude 3.7 Sonnet | 33.0 |
| Qwen3.6 Plus (2026-04-02) | 33.0 |
| Llama 3.3 70B Instruct | 34.0 |
| Claude 3.5 Sonnet (Oct 2024) | 35.0 |
| Gemini 1.5 Pro (source unspecified) | 35.0 |
| Kimi K2 Instruct | 37.0 |
| Llama 3.1 405B Instruct | 38.0 |
| Haiku 4.5 | 41.0 |
| Opus 3 | 42.0 |
| GPT-5.2 | 42.0 |
| Qwen3 Max (rolling alias) | 43.0 |
| DeepSeek-V4-Pro | 44.0 |
| Qwen3.8 Flash | 44.0 |
| DeepSeek V4.1 Flash | 45.0 |