Fiction.LiveBench - 192K tokens
Chat & Writing · 2025-02-19
Fiction.LiveBench quiz accuracy when the story is cut to about 192K tokens, so the relevant passages sit inside that much surrounding fiction. The same private, open-ended comprehension questions are asked at every length; question counts are not published.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Flash Preview | 100.0 |
| Gemini 3 Pro Preview | 96.9 |
| GPT-5.2 | 96.9 |
| Gemini 2.5 Pro | 90.6 |
| GPT-5 | 87.5 |
| Kimi K2.5 | 87.5 |
| Grok 4 | 84.4 |
| Sonoma Sky Alpha | 84.4 |
| Gemini 2.5 Flash | 78.1 |
| GPT-5.2 Pro | 78.1 |
| Grok 4 Fast | 78.1 |
| Qwen3.5 Plus (2026-02-15) | 76.0 |
| MiniMax M1 | 71.9 |
| Qwen3.6 Plus (2026-04-02) | 71.9 |
| GPT-5 Mini | 59.4 |
| o3 | 58.1 |
| GPT-4.1 | 56.3 |
| Kimi K2 Instruct 0905 | 50.0 |
| Qwen3-Next-80B-A3B-Thinking | 46.9 |
| Gemini 2.5 Flash-Lite | 43.8 |
| O4 Mini | 43.8 |
| Nemotron 3 Nano 30B A3B | 37.5 |
| GPT-5 Nano | 34.4 |
| Qwen3 Next 80B A3B Instruct | 31.3 |
| Qwen3 235B A22B Instruct 2507 | 28.6 |