Benchmark Almanac

LLM Thematic Generalization Benchmark v2

Claude Opus 4.6 80.6 leads since March 2026
Not in model cards

Models rank eight candidates using three positive examples and three counterexamples. Version 2 filters ambiguous prompts and reports a harder subset of 703 items.

The April 16, 2026 leaderboard reports Claude, GPT, Gemini and other models. Claude Opus 4.6 led the hard subset with an inverse-rank score of 80.6.

Made by
Lech Mazur
Released
March 2026
Kind
Language reasoning
Level
University
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Run by
Anyone
Strongest model tested
GPT-5.4 #28 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
24
Entries on its own leaderboard, 19 models we track

Claude Opus 4.6 leads with 80.6 among models released by April 2026 Claude Opus 4.6 leads with 80.6 among models released by April 2026

Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2026, 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 Opus 4.6 high reasoning 80.6
  2. 2 GPT-5.4 xhigh reasoning 80.0
  3. 3 Gemini 3.1 Pro 79.4
  4. 4 Claude Sonnet 4.6 high reasoning 76.3
  5. 5 Claude Opus 4.7 high reasoning 72.8
  6. 6 GLM-5.1 69.8
  7. 7 Kimi K2.5 thinking 69.4
  8. 8 Qwen3.5 397B-A17B 65.1
  9. 9 DeepSeek-V3.2 65.0
  10. 10 Grok 4.20 reasoning 63.8
  11. 11 Gemini 3.1 Flash-Lite 63.3
  12. 12 GPT-5.4 Mini xhigh reasoning 61.7
  13. 13 Qwen 3.6 Plus 59.5
  14. 14 ByteDance Seed2.0 Pro 57.1
  15. 15 Gemma 4 31B IT reasoning 53.0
  16. 16 Qwen3.5-122B-A10B 51.2
  17. 17 Xiaomi MiMo V2 Pro 45.9
  18. 18 Qwen3.5-27B 45.5
  19. 19 Baidu Ernie 5.0 41.7
  20. 20 Arcee Trinity Large Thinking 41.6
  21. 21 MiniMax-M2.7 39.3
  22. 22 Mistral Large 3 23.0
  23. 23 Grok 4.20 0309 Non-Reasoning 22.0
  24. 24 Mistral Medium 3.1 20.3

Source: Official LLM Thematic Generalization Benchmark v2 leaderboard

#EntryScoreSettingDateSource
1 Claude Opus 4.6 80.6 high reasoning Feb 2026 Official leaderboard
2 GPT-5.4 80.0 xhigh reasoning Mar 2026 Official leaderboard
3 Gemini 3.1 Pro 79.4 Feb 2026 Official leaderboard
4 Claude Sonnet 4.6 76.3 high reasoning Feb 2026 Official leaderboard
5 Claude Opus 4.7 72.8 high reasoning Apr 2026 Official leaderboard
6 GLM-5.1 69.8 Apr 2026 Official leaderboard
7 Kimi K2.5 69.4 thinking Jan 2026 Official leaderboard
8 Qwen3.5 397B-A17B 65.1 Feb 2026 Official leaderboard
9 DeepSeek-V3.2 65.0 Dec 2025 Official leaderboard
10 Grok 4.20 63.8 reasoning Feb 2026 Official leaderboard
11 Gemini 3.1 Flash-Lite 63.3 Mar 2026 Official leaderboard
12 GPT-5.4 Mini 61.7 xhigh reasoning Mar 2026 Official leaderboard
13 Qwen 3.6 Plus 59.5 Mar 2026 Official leaderboard
14 ByteDance Seed2.0 Pro 57.1 Official leaderboard
15 Gemma 4 31B IT 53.0 reasoning Apr 2026 Official leaderboard
16 Qwen3.5-122B-A10B 51.2 Feb 2026 Official leaderboard
17 Xiaomi MiMo V2 Pro 45.9 Official leaderboard
18 Qwen3.5-27B 45.5 Feb 2026 Official leaderboard
19 Baidu Ernie 5.0 41.7 Official leaderboard
20 Arcee Trinity Large Thinking 41.6 Official leaderboard
21 MiniMax-M2.7 39.3 Mar 2026 Official leaderboard
22 Mistral Large 3 23.0 Dec 2025 Official leaderboard
23 Grok 4.20 0309 22.0 Non-Reasoning Official leaderboard
24 Mistral Medium 3.1 20.3 Aug 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Opus 4.6 80.6 high reasoning Feb 2026 Official leaderboard
2 GPT-5.4 80.0 xhigh reasoning Mar 2026 Official leaderboard
3 Gemini 3.1 Pro 79.4 Feb 2026 Official leaderboard
4 Claude Sonnet 4.6 76.3 high reasoning Feb 2026 Official leaderboard
5 Claude Opus 4.7 72.8 high reasoning Apr 2026 Official leaderboard
6 GLM-5.1 69.8 Apr 2026 Official leaderboard
7 Kimi K2.5 69.4 thinking Jan 2026 Official leaderboard
8 Qwen3.5 397B-A17B 65.1 Feb 2026 Official leaderboard
9 DeepSeek-V3.2 65.0 Dec 2025 Official leaderboard
10 Grok 4.20 63.8 reasoning Feb 2026 Official leaderboard
11 Gemini 3.1 Flash-Lite 63.3 Mar 2026 Official leaderboard
12 GPT-5.4 Mini 61.7 xhigh reasoning Mar 2026 Official leaderboard
13 Qwen 3.6 Plus 59.5 Mar 2026 Official leaderboard
14 Gemma 4 31B IT 53.0 reasoning Apr 2026 Official leaderboard
15 Qwen3.5-122B-A10B 51.2 Feb 2026 Official leaderboard
16 Qwen3.5-27B 45.5 Feb 2026 Official leaderboard
17 MiniMax-M2.7 39.3 Mar 2026 Official leaderboard
18 Mistral Large 3 23.0 Dec 2025 Official leaderboard
19 Mistral Medium 3.1 20.3 Aug 2025 Official leaderboard

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 Opus 4.6
no reasoning high reason…
80.6
GPT-5.4
no reasoning medium reas…
80.0
Claude Sonnet 4.6
no reasoning high reason…
76.3
Claude Opus 4.7
no reasoning high reason…
72.8

Source: Official LLM Thematic Generalization Benchmark v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.6 high reasoning 80.6
GPT-5.4 xhigh reasoning 80.0
Gemini 3.1 Pro 79.4
GPT-5.4 high reasoning 77.4
Claude Sonnet 4.6 high reasoning 76.3
GPT-5.4 medium reasoning 75.2
Claude Opus 4.7 high reasoning 72.8
GLM-5.1 69.8
Kimi K2.5 thinking 69.4
Claude Opus 4.6 no reasoning 68.8
Claude Sonnet 4.6 no reasoning 68.5
Qwen3.5 397B-A17B 65.1
DeepSeek-V3.2 65.0
Grok 4.20 reasoning 63.8
Gemini 3.1 Flash-Lite 63.3
GPT-5.4 Mini xhigh reasoning 61.7
Qwen 3.6 Plus 59.5
Gemma 4 31B IT reasoning 53.0
Claude Opus 4.7 no reasoning 52.6
Qwen3.5-122B-A10B 51.2
Qwen3.5-27B 45.5
MiniMax-M2.7 39.3
GPT-5.4 no reasoning 27.1
Mistral Large 3 23.0
Mistral Medium 3.1 20.3

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 Opus 4.6
no reasoning high reason…
80.6
GPT-5.4
no reasoning medium reas…
80.0
Claude Sonnet 4.6
no reasoning high reason…
76.3
Claude Opus 4.7
no reasoning high reason…
72.8

Source: Official LLM Thematic Generalization Benchmark v2 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.6 high reasoning 80.6
GPT-5.4 xhigh reasoning 80.0
Gemini 3.1 Pro 79.4
GPT-5.4 high reasoning 77.4
Claude Sonnet 4.6 high reasoning 76.3
GPT-5.4 medium reasoning 75.2
Claude Opus 4.7 high reasoning 72.8
GLM-5.1 69.8
Kimi K2.5 thinking 69.4
Claude Opus 4.6 no reasoning 68.8
Claude Sonnet 4.6 no reasoning 68.5
Qwen3.5 397B-A17B 65.1
DeepSeek-V3.2 65.0
Grok 4.20 reasoning 63.8
Gemini 3.1 Flash-Lite 63.3
GPT-5.4 Mini xhigh reasoning 61.7
Qwen 3.6 Plus 59.5
Gemma 4 31B IT reasoning 53.0
Claude Opus 4.7 no reasoning 52.6
Qwen3.5-122B-A10B 51.2
Qwen3.5-27B 45.5
MiniMax-M2.7 39.3
GPT-5.4 no reasoning 27.1
Mistral Large 3 23.0
Mistral Medium 3.1 20.3

Number one over time

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

Anthropic
Own leaderboard Not tested since Apr 2026

Source: Official LLM Thematic Generalization Benchmark v2 leaderboard

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
On the benchmark’s own leaderboard
20.0 40.0 60.0 80.0 100.0 2026 20.0 40.0 60.0 80.0 100.0 2026

Source: Official LLM Thematic Generalization Benchmark v2 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 7 months ago, in April 2026, for Claude Opus 4.7. 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 official leaderboard (24 entries).
  • Pass Models scored 19 models we track, plus 5 other entries on its leaderboard.
  • Warning Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on github.com: 25 rows for models we track and 5 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