Benchmark Almanac

Flawed Fictions

Claude 3.5 Sonnet (snapshot unspecified) 0.68 tops its leaderboard Claude 3.7 Sonnet 0.66 leads the models we track
Not in model cards

Models detect plot holes in stories and identify the line responsible. The hub port uses an 83-example holdout by default.

On November 3, 2025, a GRPO-tuned Qwen2.5-7B-Instruct led the hub runs with 0.9357 mean reward on 83 examples. The paper separately compares GPT, Claude, and Llama models.

Released
April 2025
Kind
Reading comprehension
Level
University
Scored
Checked automatically
Answers
Prediction
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
o1 #118 on the capability index
Last model added
February 2025
Scores refreshed
November 20th, 2026
14
Entries in its paper, 10 models we track

Claude 3.5 Sonnet (snapshot unspecified) tops the leaderboard with 0.68 Claude 3.7 Sonnet leads with 0.66 among models released by February 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Claude 3.7 Sonnet, at 0.66. This view shows only models we track, each at its best.

The newest model with a score here came out in February 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude 3.5 Sonnet (snapshot unspecified) Verifier 0.68
  2. 2 Claude 3.7 Sonnet Extended Thinking 0.66
  3. 3 o1 Low 0.65
  4. 4 GPT-4o (snapshot unspecified) CoT 0.58
  5. 5 GPT-4-turbo (snapshot unspecified) Few-Shot 0.55
  6. 6 o3-mini Medium 0.53
  7. 7 Claude 3.5 Haiku Few-Shot 0.51
  8. 8 Llama 3.3 70B 0.38
  9. 9 DeepSeek-R1-Distill-Qwen-32B 0.35
  10. 10 DeepSeek-R1-Distill-Qwen-14B 0.33
  11. 11 GPT-4o-mini (snapshot unspecified) CoT 0.32
  12. 12 Llama 3.1-70B 0.31
  13. 13 Qwen2.5-32B 0.31
  14. 14 Llama 3.1-8B 0.10

Source: The Flawed Fictions paper

#EntryScoreSettingDateSource
1 Claude 3.5 Sonnet (snapshot unspecified) 0.68 Verifier Official leaderboard
2 Claude 3.7 Sonnet 0.66 Extended Thinking Feb 2025 Official leaderboard
3 o1 0.65 Low Dec 2024 Official leaderboard
4 GPT-4o (snapshot unspecified) 0.58 CoT Official leaderboard
5 GPT-4-turbo (snapshot unspecified) 0.55 Few-Shot Official leaderboard
6 o3-mini 0.53 Medium Jan 2025 Official leaderboard
7 Claude 3.5 Haiku 0.51 Few-Shot Oct 2024 Official leaderboard
8 Llama 3.3 70B 0.38 Dec 2024 Official leaderboard
9 DeepSeek-R1-Distill-Qwen-32B 0.35 Jan 2025 Official leaderboard
10 DeepSeek-R1-Distill-Qwen-14B 0.33 Jan 2025 Official leaderboard
11 GPT-4o-mini (snapshot unspecified) 0.32 CoT Official leaderboard
12 Llama 3.1-70B 0.31 Jul 2024 Official leaderboard
13 Qwen2.5-32B 0.31 Sep 2024 Official leaderboard
14 Llama 3.1-8B 0.10 Jul 2024 Official leaderboard
#ModelScoreSettingDateSource
1 Claude 3.7 Sonnet 0.66 Extended Thinking Feb 2025 Official leaderboard
2 o1 0.65 Low Dec 2024 Official leaderboard
3 o3-mini 0.53 Medium Jan 2025 Official leaderboard
4 Claude 3.5 Haiku 0.51 Few-Shot Oct 2024 Official leaderboard
5 Llama 3.3 70B 0.38 Dec 2024 Official leaderboard
6 DeepSeek-R1-Distill-Qwen-32B 0.35 Jan 2025 Official leaderboard
7 DeepSeek-R1-Distill-Qwen-14B 0.33 Jan 2025 Official leaderboard
8 Llama 3.1-70B 0.31 Jul 2024 Official leaderboard
9 Qwen2.5-32B 0.31 Sep 2024 Official leaderboard
10 Llama 3.1-8B 0.10 Jul 2024 Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s paper, each ranking models on one part of the test, such as Accuracy, Precision and Recall. Claude 3.7 Sonnet leads 3 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Accuracy

10 models we track have a score here.

  1. 1 Claude 3.7 Sonnet 0.73
  2. 2 o1 0.71
  3. 3 o3-mini 0.63
  4. 4 DeepSeek-R1-Distill-Qwen-14B ↑3 0.58
  5. 5 Claude 3.5 Haiku ↓1 0.57
  6. 6 Llama 3.3 70B ↓1 0.57
  7. 7 Llama 3.1-70B ↑1 0.56
  8. 8 DeepSeek-R1-Distill-Qwen-32B ↓2 0.56
  9. 9 Qwen2.5-32B 0.53
  10. 10 Llama 3.1-8B 0.50

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3.7 Sonnet
Extended Th…
0.66
o1
Low
0.65
o3-mini
Low High
0.53
Claude 3.5 Haiku
CoT Few-Shot
0.51

Source: The Flawed Fictions paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3.7 Sonnet Extended Thinking 0.66
o1 Low 0.65
o1 Medium 0.65
o1 High 0.64
Claude 3.7 Sonnet 0.55
o3-mini Medium 0.53
o3-mini Low 0.52
Claude 3.5 Haiku Few-Shot 0.51
o3-mini High 0.47
Claude 3.5 Haiku 0.46
Claude 3.5 Haiku CoT 0.46
Llama 3.3 70B 0.38
DeepSeek-R1-Distill-Qwen-32B 0.35
DeepSeek-R1-Distill-Qwen-14B 0.33
Llama 3.1-70B 0.31
Qwen2.5-32B 0.31
Llama 3.1-8B 0.10

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3.7 Sonnet
Extended Th…
0.66
o1
Low
0.65
o3-mini
Low High
0.53
Claude 3.5 Haiku
CoT Few-Shot
0.51

Source: The Flawed Fictions paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude 3.7 Sonnet Extended Thinking 0.66
o1 Low 0.65
o1 Medium 0.65
o1 High 0.64
Claude 3.7 Sonnet 0.55
o3-mini Medium 0.53
o3-mini Low 0.52
Claude 3.5 Haiku Few-Shot 0.51
o3-mini High 0.47
Claude 3.5 Haiku 0.46
Claude 3.5 Haiku CoT 0.46
Llama 3.3 70B 0.38
DeepSeek-R1-Distill-Qwen-32B 0.35
DeepSeek-R1-Distill-Qwen-14B 0.33
Llama 3.1-70B 0.31
Qwen2.5-32B 0.31
Llama 3.1-8B 0.10

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Benchmark paper Not tested since Apr 2025

Source: The Flawed Fictions paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper
0.00 0.20 0.40 0.60 0.80 2025 0.00 0.20 0.40 0.60 0.80 2025

Source: Official Flawed Fictions leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Knowledge

All 332 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy. Seed1.8 92.3%
  • SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (14 entries).
  • Pass Models scored 10 models we track, plus 4 other entries on its leaderboard.
  • Warning Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o1 (#118).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 17 rows for models we track and 4 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check