Benchmark Almanac

AstroMLab 1

Claude-3.5-Sonnet (snapshot unspecified) 85% tops its leaderboard Claude 3 Opus 82.7% leads the models we track
Not in model cards

Models answer 4,425 multiple choice questions drawn from astronomy review articles. The authors publish the questions and grade submissions with a private answer key.

The AstroMLab 4 paper reports 86.2% accuracy for AstroSage on the original questions and compares it with general purpose models.

Made by
AstroMLab
Released
July 2024
Kind
Science questions
Level
Expert
Scored
Checked automatically
Answers
Multiple choice
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
A host
Strongest model tested
Claude 3 Opus #173 on the capability index
Last model added
September 2024
Scores refreshed
November 20th, 2026
59
Entries in its paper, 24 models we track

Claude-3.5-Sonnet (snapshot unspecified) tops the leaderboard with 85% Claude 3 Opus leads with 82.7% among models released by September 2024

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. 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 Opus, at 82.7%. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2024, 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) 85%
  2. 2 Claude 3 Opus 82.7%
  3. 3 Llama 3-70B 80.6%
  4. 4 GPT-4o (snapshot unspecified) 80.4%
  5. 5 Claude 3 Haiku 77.9%
  6. 6 Mixtral-8x22B-v0.1 77.7%
  7. 7 Qwen-2-70B 77.7%
  8. 8 Gemini-1.5-Pro (snapshot unspecified) 77.6%
  9. 9 Yi-Large 77.3%
  10. 10 Claude 3 Sonnet 76.7%
  11. 11 Step-2 76.6%
  12. 12 Phi-3-14B 75.6%
  13. 13 Gemma 2 27B 75.6%
  14. 14 Claude 2 75.3%
  15. 15 Step-1 75.2%
  16. 16 ERNIE-4.0 75.1%
  17. 17 GLM-4-0520 75.1%
  18. 18 GPT-4 (snapshot unspecified) 74.5%
  19. 19 Mistral-8x7B-v0.1 73.7%
  20. 20 Deepseek-v2 73.6%
  21. 21 Gemini-1.5-Flash (snapshot unspecified) 73.6%
  22. 22 Qwen1.5-32B 73.2%
  23. 23 Yi-1.5-34B 73.1%
  24. 24 GLM-4-Air 72.9%
  25. 25 Llama 3-8B 72.9%
  26. 26 ABAB-6.5 72.7%
  27. 27 Qwen1.5-110B 72.7%
  28. 28 GLM-4-AirX 72.5%
  29. 29 Moonshot-v1 72.3%
  30. 30 ERNIE-3.5 72.1%
  31. 31 Qwen2.5-7B 71.8%
  32. 32 Phi-3-4B 71.7%
  33. 33 Gemma 2 9B 71.5%
  34. 34 Gemini 1.0 Pro 71%
  35. 35 Llama 2-70B 70.7%
  36. 36 GPT-3.5 (snapshot unspecified) 70.4%
  37. 37 Yi-Medium 70.3%
  38. 38 Doubao-Pro 70.1%
  39. 39 ABAB-5.5 69.5%
  40. 40 Yi-1.5-9B 68.4%
  41. 41 Qwen2-7B 68%
  42. 42 Qwen1.5-14B 67.7%
  43. 43 GLM-4-Flash 67.1%
  44. 44 GLM-4-9B 67%
  45. 45 Phi-2-3B 65.6%
  46. 46 InternLM-2.5-7B 64.5%
  47. 47 GLM-3-Turbo 64.3%
  48. 48 Mistral 7B v0.3 63.9%
  49. 49 Qwen1.5-7B 63.7%
  50. 50 Deepseek-67B 63.1%
  51. 51 Mistral 7B v0.2 62.1%
  52. 52 Yi-1.5-6B 61%
  53. 53 Doubao-Lite 60.5%
  54. 54 Qwen-1-7B 57.4%
  55. 55 Gemma-1-7B 56.1%
  56. 56 ChatGLM3-6B 50.4%
  57. 57 Llama 2-7B 50.3%
  58. 58 Mistral 7B v0.1 48.1%
  59. 59 Gemma-1-2B 44.1%

Source: The AstroMLab 1 paper

#EntryScoreSettingDateSource
1 Claude-3.5-Sonnet (snapshot unspecified) 85% Official leaderboard
2 Claude 3 Opus 82.7% Feb 2024 Official leaderboard
3 Llama 3-70B 80.6% Apr 2024 Official leaderboard
4 GPT-4o (snapshot unspecified) 80.4% Official leaderboard
5 Claude 3 Haiku 77.9% Mar 2024 Official leaderboard
6 Mixtral-8x22B-v0.1 77.7% Official leaderboard
7 Qwen-2-70B 77.7% Official leaderboard
8 Gemini-1.5-Pro (snapshot unspecified) 77.6% Official leaderboard
9 Yi-Large 77.3% May 2024 Official leaderboard
10 Claude 3 Sonnet 76.7% Feb 2024 Official leaderboard
11 Step-2 76.6% Official leaderboard
12 Phi-3-14B 75.6% Official leaderboard
13 Gemma 2 27B 75.6% Jun 2024 Official leaderboard
14 Claude 2 75.3% Jul 2023 Official leaderboard
15 Step-1 75.2% Official leaderboard
16 ERNIE-4.0 75.1% Official leaderboard
17 GLM-4-0520 75.1% Official leaderboard
18 GPT-4 (snapshot unspecified) 74.5% Official leaderboard
19 Mistral-8x7B-v0.1 73.7% Official leaderboard
20 Deepseek-v2 73.6% Official leaderboard
21 Gemini-1.5-Flash (snapshot unspecified) 73.6% Official leaderboard
22 Qwen1.5-32B 73.2% Feb 2024 Official leaderboard
23 Yi-1.5-34B 73.1% May 2024 Official leaderboard
24 GLM-4-Air 72.9% Official leaderboard
25 Llama 3-8B 72.9% Apr 2024 Official leaderboard
26 ABAB-6.5 72.7% Official leaderboard
27 Qwen1.5-110B 72.7% Apr 2024 Official leaderboard
28 GLM-4-AirX 72.5% Official leaderboard
29 Moonshot-v1 72.3% Official leaderboard
30 ERNIE-3.5 72.1% Official leaderboard
31 Qwen2.5-7B 71.8% Sep 2024 Official leaderboard
32 Phi-3-4B 71.7% Official leaderboard
33 Gemma 2 9B 71.5% Jun 2024 Official leaderboard
34 Gemini 1.0 Pro 71% Dec 2023 Official leaderboard
35 Llama 2-70B 70.7% Jul 2023 Official leaderboard
36 GPT-3.5 (snapshot unspecified) 70.4% Official leaderboard
37 Yi-Medium 70.3% Official leaderboard
38 Doubao-Pro 70.1% Official leaderboard
39 ABAB-5.5 69.5% Official leaderboard
40 Yi-1.5-9B 68.4% Official leaderboard
41 Qwen2-7B 68% Jun 2024 Official leaderboard
42 Qwen1.5-14B 67.7% Feb 2024 Official leaderboard
43 GLM-4-Flash 67.1% Official leaderboard
44 GLM-4-9B 67% Jun 2024 Official leaderboard
45 Phi-2-3B 65.6% Official leaderboard
46 InternLM-2.5-7B 64.5% Official leaderboard
47 GLM-3-Turbo 64.3% Official leaderboard
48 Mistral 7B v0.3 63.9% Sep 2023 Official leaderboard
49 Qwen1.5-7B 63.7% Feb 2024 Official leaderboard
50 Deepseek-67B 63.1% Official leaderboard
51 Mistral 7B v0.2 62.1% Dec 2023 Official leaderboard
52 Yi-1.5-6B 61% Official leaderboard
53 Doubao-Lite 60.5% Official leaderboard
54 Qwen-1-7B 57.4% Official leaderboard
55 Gemma-1-7B 56.1% Official leaderboard
56 ChatGLM3-6B 50.4% Oct 2023 Official leaderboard
57 Llama 2-7B 50.3% Jul 2023 Official leaderboard
58 Mistral 7B v0.1 48.1% Sep 2023 Official leaderboard
59 Gemma-1-2B 44.1% Official leaderboard
#ModelScoreSettingDateSource
1 Claude 3 Opus 82.7% Feb 2024 Official leaderboard
2 Llama 3-70B 80.6% Apr 2024 Official leaderboard
3 Claude 3 Haiku 77.9% Mar 2024 Official leaderboard
4 Yi-Large 77.3% May 2024 Official leaderboard
5 Claude 3 Sonnet 76.7% Feb 2024 Official leaderboard
6 Gemma 2 27B 75.6% Jun 2024 Official leaderboard
7 Claude 2 75.3% Jul 2023 Official leaderboard
8 Qwen1.5-32B 73.2% Feb 2024 Official leaderboard
9 Yi-1.5-34B 73.1% May 2024 Official leaderboard
10 Llama 3-8B 72.9% Apr 2024 Official leaderboard
11 Qwen1.5-110B 72.7% Apr 2024 Official leaderboard
12 Qwen2.5-7B 71.8% Sep 2024 Official leaderboard
13 Gemma 2 9B 71.5% Jun 2024 Official leaderboard
14 Gemini 1.0 Pro 71% Dec 2023 Official leaderboard
15 Llama 2-70B 70.7% Jul 2023 Official leaderboard
16 Qwen2-7B 68% Jun 2024 Official leaderboard
17 Qwen1.5-14B 67.7% Feb 2024 Official leaderboard
18 GLM-4-9B 67% Jun 2024 Official leaderboard
19 Mistral 7B v0.3 63.9% Sep 2023 Official leaderboard
20 Qwen1.5-7B 63.7% Feb 2024 Official leaderboard
21 Mistral 7B v0.2 62.1% Dec 2023 Official leaderboard
22 ChatGLM3-6B 50.4% Oct 2023 Official leaderboard
23 Llama 2-7B 50.3% Jul 2023 Official leaderboard
24 Mistral 7B v0.1 48.1% Sep 2023 Official leaderboard

Rankings by part

We read 11 parts from the benchmark’s paper in 2 groups: subfield and tested ability. Each ranks models on one part of the test. Claude 3 Opus leads 10 of them, and Llama 3-70B leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

Solar & Stellar

24 models we track have a score here.

  1. 1 Claude 3 Opus 81.5%
  2. 2 Llama 3-70B 78.7%
  3. 3 Claude 3 Haiku 77.6%
  4. 4 Claude 3 Sonnet ↑1 76.1%
  5. 5 Yi-Large ↓1 75.6%
  6. 6 Claude 2 ↑1 74.1%
  7. 7 Gemma 2 27B ↓1 73.3%
  8. 8 Llama 3-8B ↑2 71.2%
  9. 9 Qwen1.5-110B ↑2 71.2%
  10. 10 Gemma 2 9B ↑3 71%

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 Sep 2024

Source: The AstroMLab 1 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 Best so far
0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Claude 3 Opus Claude 2 0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Claude 3 Opus Claude 2

Source: Official AstroMLab 1 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 Science and reasoning

All 203 in Science and reasoning
  • GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics. Gemini 4 Carbon 97%
  • Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
  • BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
  • FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics. Intern-S2-397B 87%
  • HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels. Gemini 3 Pro 78.3%
  • LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in September 2024, for Qwen2.5-7B. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (59 entries).
  • Pass Models scored 24 models we track, plus 35 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-7B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude 3 Opus (#173).
  • Pass Headroom The best is 82.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 24 rows for models we track and 35 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