Fiction.LiveBench 120k
Long-context comprehension of fiction at 120k tokens.
As of 25 Oct 2026, o3 leads Fiction.LiveBench 120k on BenchLeader with 100.0%, ahead of Grok 4 at 96.9%, across 35 model configurations with a published result.
- Published by
- Fiction.livedata via Epoch AI Benchmarking Hub
- Category
- Long context
- Index weight
- 1.0
- Models
- 35
- Data as of
- 25 Oct 2026
- Newest result
- 27 Jan 2026
The publisher has posted no new results on this board since 27 Jan 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
Fiction.LiveBench tests reading comprehension over long stories: the model is given up to 120,000 tokens of fiction and asked questions whose answers require tracking characters and events across the whole text.
How it is scored
Percent of questions answered correctly at the 120k-token context length.
What to keep in mind
Long-context scores fall off sharply with length for most models, so the 120k figure can differ a lot from shorter-context results. The stories are from a fan-fiction site and may be in training data.
- 1o3 (medium)100.0%
- 2Grok 496.9%
- 3GPT-5 (medium)96.9%
- 4Gemini 2.5 Pro90.6%
- 5o3-pro (medium)88.9%
- 6Kimi K2.578.1%
- 7Grok 4 Fast75.0%
- 8Gemini 2.5 Flash68.8%
- 9Qwen3 235B A22B 2507 (thinking)68.8%
- 10Chatgpt 4o 01.29 202565.6%
- 11Grok 3 mini (medium)65.6%
- 12GPT-4.563.9%
- 13GPT-5 mini (medium)62.5%
- 14o4-mini (medium)62.5%
- 15GPT-4.162.5%
| # | ||||
|---|---|---|---|---|
| 1 |  o3medium | 100.0% | 53.3 | 2025-04-16 |
| 2 |  Grok 4 | 96.9% | 58.0 | 2025-07-09 |
| 3 | GPT-5medium | 96.9% | 57.5 | 2025-08-07 |
| 4 |  Gemini 2.5 Pro | 90.6% | 53.9 | 2025-03-25 |
| 5 | o3-promedium | 88.9% | – | 2025-06-10 |
| 6 |  Kimi K2.5open ↗ | 78.1% | 54.5 | 2026-01-27 |
| 7 | Grok 4 Fast | 75.0% | – | 2025-09-19 |
| 8 | Gemini 2.5 Flash | 68.8% | 52.7 | 2025-05-20 |
| 9 |  Qwen3 235B A22B 2507thinkingopen ↗ | 68.8% | 50.3 | 2025-07-25 |
| 10 | Chatgpt 4o 01.29 2025 | 65.6% | – | 2025-01-29 |
| 11 | Grok 3 minimedium | 65.6% | – | 2025-04-09 |
| 12 | GPT-4.5 | 63.9% | 49.7 | 2025-02-27 |
| 13 | GPT-5 minimedium | 62.5% | 53.8 | 2025-08-07 |
| 14 | o4-minimedium | 62.5% | 51.9 | 2025-04-16 |
| 15 | GPT-4.1 | 62.5% | 48.3 | 2025-04-14 |
| 16 | Qwen3-Next 80B A3Bopen ↗ | 62.5% | 48.3 | 2025-09-11 |
| 17 | Gemini 2.0 Flash | 62.5% | 43.2 | 2025-02-05 |
| 18 |  MiniMax-M1open | 59.4% | 47.5 | 2025-06-13 |
| 19 | Grok 3 | 58.3% | 51.5 | 2025-04-09 |
| 20 |  Claude 3.7 Sonnet | 53.1% | 50.0 | 2025-02-24 |
| 21 | o1medium | 53.1% | 48.1 | 2024-12-17 |
| 22 | GPT-4.1 mini | 46.9% | 45.0 | 2025-04-14 |
| 23 | Qwen3 235B A22B (Parasail) | 44.4% | – | 2025-09-01 |
| 24 | o3-minimedium | 43.8% | 45.2 | 2025-01-31 |
| 25 | Kimi K2 0905open ↗ | 40.6% | 50.5 | 2025-09-05 |
| 26 | Claude Opus 4 | 37.5% | 52.6 | 2025-05-22 |
| 27 | Gemini 2.0 Pro | 37.5% | 47.0 | 2025-02-05 |
| 28 | Gemini 2.0 Flash Thinkingthinking | 37.5% | 44.9 | 2025-01-21 |
| 29 | Claude Sonnet 4 | 36.4% | 50.5 | 2025-05-22 |
| 30 |  Llama 4 Maverickopen ↗ | 36.4% | 42.8 | 2025-04-06 |
| 31 |  DeepSeek R1open ↗ | 33.3% | 46.8 | 2025-01-20 |
| 32 | Llama 4 Scoutopen ↗ | 27.3% | 39.7 | 2025-04-06 |
| 33 | Gemini 2.5 Flash-Litethinking | 21.9% | 46.0 | 2025-06-17 |
| 34 | GPT-5 nanomedium | 21.9% | 45.9 | 2025-08-07 |
| 35 | GPT-4.1 nano | 18.8% | 38.0 | 2025-04-14 |
More long context benchmarks · All LLM benchmarks
Cite as: BenchLeader, “Fiction.LiveBench 120k leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/fiction_livebench.
Fiction.LiveBench 120k: questions
- What does Fiction.LiveBench 120k measure?
- Fiction.LiveBench tests reading comprehension over long stories: the model is given up to 120,000 tokens of fiction and asked questions whose answers require tracking characters and events across the whole text. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads Fiction.LiveBench 120k?
- o3 leads Fiction.LiveBench 120k with 100.0% as of 25 Oct 2026, ahead of Grok 4 at 96.9%.
- How many models have Fiction.LiveBench 120k results?
- 35 model configurations have a Fiction.LiveBench 120k result on BenchLeader, all taken from Fiction.live via Epoch AI Benchmarking Hub.
- Who runs Fiction.LiveBench 120k and how often is it updated?
- Fiction.LiveBench 120k is published by Fiction.live. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does Fiction.LiveBench 120k count toward the BenchLeader Index?
- Yes. Fiction.LiveBench 120k contributes to the long context category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.