Fiction.LiveBench - 60K tokens
Chat & Writing · 2025-02-19
Fiction.LiveBench quiz accuracy when the story is cut to about 60K 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. This is the family's preferred observation: the longest length nearly every evaluated model reports.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3 Pro Preview | 100.0 |
| GPT-5 | 100.0 |
| Opus 4.5 | 97.2 |
| GPT-5.2 | 97.2 |
| Grok 4 | 97.2 |
| Gemini 3 Flash Preview | 96.9 |
| Sonnet 4 | 94.4 |
| Opus 4.6 | 91.7 |
| Qwen3.5 Plus (2026-02-15) | 91.2 |
| Sonnet 4.5 | 88.9 |
| Kimi K2.5 | 88.9 |
| GLM-5 | 87.5 |
| Gemini 2.5 Pro | 83.3 |
| o3 | 83.3 |
| Sonoma Sky Alpha | 83.3 |
| GLM-4.7 | 80.6 |
| GPT-5.2 Pro | 80.6 |
| Grok 4 Fast | 80.6 |
| Qwen3 235B A22B Thinking 2507 | 77.8 |
| DeepSeek-V3.2-Exp | 71.4 |
| Opus 4 | 69.4 |
| GPT-4.5 | 66.7 |
| O4 Mini | 66.7 |
| Qwen3.6 Plus (2026-04-02) | 66.7 |
| DeepSeek-V3.2 | 63.9 |
| Gemini 2.5 Flash | 63.9 |
| GPT-5 Mini | 61.1 |
| QwQ-32B | 61.1 |
| DeepSeek-R1-0528 | 58.3 |
| DeepSeek-V3.1 | 58.3 |
| MiniMax M1 | 58.3 |
| Gemini 2.5 Flash Preview 04-17 | 58.3 |
| ChatGPT-4o Latest (source-unspecified snapshot) | 55.6 |
| GPT-5 Chat | 55.6 |
| Kimi K2 Instruct 0905 | 55.6 |
| Qwen3-Next-80B-A3B-Thinking | 55.6 |
| GPT-4.1 | 52.8 |
| MiniMax M2.5 | 50.0 |
| Qwen3 235B A22B Instruct 2507 | 50.0 |
| GLM-4.5 | 47.2 |