Benchmark Almanac

Furniture Assembly Benchmark

Claude Opus 5.5 83.3% leads since Sep 2026 Not in model cards

Model gets an IKEA manual and a photo of a partly built item, then names the faulty step or next step. Accuracy on 60 photos.

LLM grader checks error descriptions. Novice human baseline 63%. Photos and manuals are not distributed.

Made by Epoch AI Released Sep 2026 Spatial reasoning and counting Tasks kept private Last model added Oct 2026 Scores refreshed 31 Oct 2026
Level
Everyday Most adults, without training
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Short answer A number, word or short string matched to a key.
Paper Leaderboard
35
Models scored in Epoch AI data

Claude Opus 5.5 leads with 83.3%

Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.

35 models. Scroll sideways for the rest.

  1. 83.3 Claude Opus 5.5
  2. 80.0 GPT-6 Astra
  3. 80.0 GPT-6.1 Sol
  4. 79.2 GPT-6.1 Terra
  5. 75.0 Claude Sonnet 5.5
  6. 72.5 Gemini 4 Argon
  7. 70.0 Claude Fable 5.1
  8. 60.8 Claude Opus 5
  9. 58.3 GPT-6 Sol
  10. 56.7 GPT-5.6 Sol
  11. 54.2 GPT-5.6 Terra
  12. 54.2 Muse Wonder
  13. 47.5 Claude Haiku 5.5
  14. 44.2 GPT-5.5
  15. 44.2 GPT-6 Luna
  16. 42.5 Claude Opus 4.8
  17. 42.5 GPT-5.6 Luna
  18. 40.0 Grok 4.6
  19. 38.3 GPT-5.2
  20. 37.5 GPT-5.4
  21. 35.8 Claude Fable 5
  22. 34.2 Kimi K3
  23. 34.2 DeepSeek V4.1 Flash
  24. 33.3 Claude Opus 4.7
  25. 31.7 Gemini 3.8 Flash
  26. 28.3 Claude Opus 4.5
  27. 28.3 Claude Opus 4.6
  28. 26.7 Gemini 3.1 Pro
  29. 26.7 Gemini 3.7 Flash
  30. 23.3 Gemini 3.6 Flash
  31. 22.5 Grok 4.5
  32. 21.7 Kimi K2.6
  33. 20.8 Grok 4.7
  34. 20.8 Mistral Large 4
  35. 20.0 Qwen3.8 Max (0902)
  1. 1 Claude Opus 5.5 max 83.3%
  2. 2 GPT-6 Astra max 80%
  3. 3 GPT-6.1 Sol max 80%
  4. 4 GPT-6.1 Terra max 79.2%
  5. 5 Claude Sonnet 5.5 max 75%
  6. 6 Gemini 4 Argon high 72.5%
  7. 7 Claude Fable 5.1 max 70%
  8. 8 Claude Opus 5 max 60.8%
  9. 9 GPT-6 Sol max 58.3%
  10. 10 GPT-5.6 Sol max 56.7%
  11. 11 GPT-5.6 Terra max 54.2%
  12. 12 Muse Wonder max 54.2%
  13. 13 Claude Haiku 5.5 max 47.5%
  14. 14 GPT-5.5 xhigh 44.2%
  15. 15 GPT-6 Luna max 44.2%
  16. 16 Claude Opus 4.8 max 42.5%
  17. 17 GPT-5.6 Luna max 42.5%
  18. 18 Grok 4.6 xhigh 40%
  19. 19 GPT-5.2 xhigh 38.3%
  20. 20 GPT-5.4 xhigh 37.5%
  21. 21 Claude Fable 5 max 35.8%
  22. 22 Kimi K3 max 34.2%
  23. 23 DeepSeek V4.1 Flash max 34.2%
  24. 24 Claude Opus 4.7 max 33.3%
  25. 25 Gemini 3.8 Flash high 31.7%
  26. 26 Claude Opus 4.5 64k thinking 28.3%
  27. 27 Claude Opus 4.6 max 28.3%
  28. 28 Gemini 3.1 Pro high 26.7%
  29. 29 Gemini 3.7 Flash high 26.7%
  30. 30 Gemini 3.6 Flash high 23.3%
  31. 31 Grok 4.5 high 22.5%
  32. 32 Kimi K2.6 21.7%
  33. 33 Grok 4.7 xhigh 20.8%
  34. 34 Mistral Large 4 high 20.8%
  35. 35 Qwen3.8 Max (0902) xhigh 20%
#ModelScoreSettingDateSource
1 Claude Opus 5.5 83.3% max Sep 2026 Epoch AI
2 GPT-6 Astra 80% max Sep 2026 Epoch AI
3 GPT-6.1 Sol 80% max Sep 2026 Epoch AI
4 GPT-6.1 Terra 79.2% max Oct 2026 Epoch AI
5 Claude Sonnet 5.5 75% max Sep 2026 Epoch AI
6 Gemini 4 Argon 72.5% high Sep 2026 Epoch AI
7 Claude Fable 5.1 70% max Sep 2026 Epoch AI
8 Claude Opus 5 60.8% max Jul 2026 Epoch AI
9 GPT-6 Sol 58.3% max Sep 2026 Epoch AI
10 GPT-5.6 Sol 56.7% max Jul 2026 Epoch AI
11 GPT-5.6 Terra 54.2% max Jul 2026 Epoch AI
12 Muse Wonder 54.2% max Oct 2026 Epoch AI
13 Claude Haiku 5.5 47.5% max Oct 2026 Epoch AI
14 GPT-5.5 44.2% xhigh Apr 2026 Epoch AI
15 GPT-6 Luna 44.2% max Sep 2026 Epoch AI
16 Claude Opus 4.8 42.5% max May 2026 Epoch AI
17 GPT-5.6 Luna 42.5% max Jul 2026 Epoch AI
18 Grok 4.6 40% xhigh Aug 2026 Epoch AI
19 GPT-5.2 38.3% xhigh Dec 2025 Epoch AI
20 GPT-5.4 37.5% xhigh Mar 2026 Epoch AI
21 Claude Fable 5 35.8% max Jun 2026 Epoch AI
22 Kimi K3 34.2% max Jul 2026 Epoch AI
23 DeepSeek V4.1 Flash 34.2% max Sep 2026 Epoch AI
24 Claude Opus 4.7 33.3% max Apr 2026 Epoch AI
25 Gemini 3.8 Flash 31.7% high Sep 2026 Epoch AI
26 Claude Opus 4.5 28.3% 64k thinking Nov 2025 Epoch AI
27 Claude Opus 4.6 28.3% max Feb 2026 Epoch AI
28 Gemini 3.1 Pro 26.7% high Feb 2026 Epoch AI
29 Gemini 3.7 Flash 26.7% high Aug 2026 Epoch AI
30 Gemini 3.6 Flash 23.3% high Jul 2026 Epoch AI
31 Grok 4.5 22.5% high Jul 2026 Epoch AI
32 Kimi K2.6 21.7% Apr 2026 Epoch AI
33 Grok 4.7 20.8% xhigh Sep 2026 Epoch AI
34 Mistral Large 4 20.8% high Oct 2026 Epoch AI
35 Qwen3.8 Max (0902) 20% xhigh Sep 2026 Epoch AI

3 models have held number one since its release in Sep 2026

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

Anthropic OpenAI
Epoch AI runs 3 leaders
Claude Opus 5.5 83.3%

Scores over time

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

Run or collected by Epoch AI Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Random guessing Claude Opus 5.5 Claude Fable 5.1 Claude Opus 5 GPT-5.6 Sol GPT-5.5 GPT-5.2 Claude Opus 4.5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Random guessing Claude Opus 5.5 Claude Fable 5.1

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Vision

All 234 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. GPT-5.1 85.4%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. Kimi K3.1 Pro 84.1%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • ChartQA In use Models answer human-written and generated questions about chart images that need visual and arithmetic reasoning. Scored by relaxed accuracy with a 5 percent numeric tolerance. Command A Vision 90.9%
  • DocVQA In use Models answer 50,000 questions about 12,000 scanned document images by reading their text and layout. Scoring is ANLS, a string similarity to reference answers. Seed1.5-VL 96.9%