AstroVLBench
Models interpret galaxy images, light curves, and spectra across five observing tasks. The authors compare their classifications with reference labels and examine their explanations.
The AstraQ-VL team independently scored class labels on the optical and radio imaging tasks. The dataset host requires users to accept access terms.
- Made by
- Wenke Ren and colleagues
- Released
- April 2026
- Kind
- Lab and data work
- Level
- University
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Subject
- Physical sciences
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Vision
- Strongest model tested
- GPT-5.2 #51 on the capability index
- Last model added
- February 2026
- Scores refreshed
- November 20th, 2026
Gemini 3 Pro leads with 68.8% among models released by February 2026
Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall accuracy, so this is the plain mean of eight accuracies across five observing tasks. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in February 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Gemini 3 Pro 68.8%
- 2 GPT-5.2 59.3%
- 3
Intern-S1-Pro 51.1%
- 4 Claude Opus 4.5 50.6%
- 5 Grok 4 49.5%
- 6
Qwen3-235B-A22B 41.4%
Source: The AstroVLBench paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 68.8% | Nov 2025 | Official leaderboard | |
| 2 | GPT-5.2 | 59.3% | Dec 2025 | Official leaderboard | |
| 3 | | 51.1% | Feb 2026 | Official leaderboard | |
| 4 | Claude Opus 4.5 | 50.6% | Nov 2025 | Official leaderboard | |
| 5 | Grok 4 | 49.5% | Jul 2025 | Official leaderboard | |
| 6 | | 41.4% | Apr 2025 | Official leaderboard |
Rankings by part
We read 8 parts from the benchmark’s paper, each ranking models on one part of the test, such as QSO host galaxy classification, Radio morphology: FIRST and Radio morphology: NVSS. Gemini 3 Pro leads 7 of them, and Claude Opus 4.5 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
QSO host galaxy classification
6 models we track have a score here.
- 1
Gemini 3 Pro 74.5%
- 2 GPT-5.2 65.2%
- 3 Grok 4 ↑2 63.6%
- 4
Intern-S1-Pro ↓1 58.2%
- 5
Qwen3-235B-A22B ↑1 46.1%
- 6 Claude Opus 4.5 ↓2 46.1%
Radio morphology: FIRST
6 models we track have a score here.
- 1
Gemini 3 Pro 84.6%
- 2 Grok 4 ↑3 70.5%
- 3 GPT-5.2 ↓1 66.1%
- 4
Intern-S1-Pro ↓1 65.1%
- 5
Qwen3-235B-A22B ↑1 52.3%
- 6 Claude Opus 4.5 ↓2 47.2%
Radio morphology: NVSS
6 models we track have a score here.
- 1
Gemini 3 Pro 55.3%
- 2
Qwen3-235B-A22B ↑4 52.3%
- 3 GPT-5.2 ↓1 50.1%
- 4 Grok 4 ↑1 49.7%
- 5
Intern-S1-Pro ↓2 48.5%
- 6 Claude Opus 4.5 ↓2 46.6%
SED classification
6 models we track have a score here.
- 1
Gemini 3 Pro 57.7%
- 2 GPT-5.2 54.2%
- 3 Claude Opus 4.5 ↑1 53.6%
- 4
Intern-S1-Pro ↓1 33.3%
- 5 Grok 4 32.1%
- 6
Qwen3-235B-A22B 26.2%
Light curve classification
6 models we track have a score here.
- 1
Gemini 3 Pro 61.3%
- 2 GPT-5.2 38.7%
- 3 Grok 4 ↑2 31%
- 4
Intern-S1-Pro ↓1 31%
- 5
Qwen3-235B-A22B ↑1 21.1%
- 6 Claude Opus 4.5 ↓2 21.1%
Spectral interpretation: Q1
6 models we track have a score here.
- 1 Claude Opus 4.5 ↑3 82%
- 2 GPT-5.2 75.6%
- 3 Grok 4 ↑2 75.4%
- 4
Gemini 3 Pro ↓3 73.9%
- 5
Intern-S1-Pro ↓2 66.4%
- 6
Qwen3-235B-A22B 28.6%
Spectral interpretation: Q2
6 models we track have a score here.
- 1
Gemini 3 Pro 87.4%
- 2 Claude Opus 4.5 ↑2 80.6%
- 3
Intern-S1-Pro 80.2%
- 4
Qwen3-235B-A22B ↑2 80%
- 5 GPT-5.2 ↓3 79.8%
- 6 Grok 4 ↓1 50%
Spectral interpretation: Q3
6 models we track have a score here.
- 1
Gemini 3 Pro 55.8%
- 2 GPT-5.2 44.5%
- 3 Claude Opus 4.5 ↑1 27.8%
- 4
Intern-S1-Pro ↓1 26%
- 5
Qwen3-235B-A22B ↑1 24.8%
- 6 Grok 4 ↓1 23.8%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The AstroVLBench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official AstroVLBench 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
- 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 9 months ago, in February 2026, for Intern-S1-Pro. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (6 entries).
- Pass Models scored 6 models we track.
- Warning Newest score The newest came 9 months ago, in February 2026, for Intern-S1-Pro.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.2 (#51).
- Pass Headroom The best is 68.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track 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.