Benchmark Almanac

SciTaste-Bench

Qwen 3.8 Max 28.0 leads since October 2026
Not in model cards

Models judge the research value of papers in six disciplines. The benchmark compares their judgments with a reference index built from seven dimensions.

Released
October 2026
Kind
Research science
Level
Expert
Scored
Checked automatically
Answers
Prediction
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text, Actions
Interaction
Agentic
Environment
APIs and tools, Browser
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Claude Fable 5 #12 on the capability index
Last model added
August 2026
Scores refreshed
November 20th, 2026
7
Models scored on its own leaderboard

Qwen 3.8 Max leads with 28.0 among models released by August 2026

Best score per model 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.

The newest model with a score here came out in August 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 Qwen 3.8 Max ADS tools 28.0
  2. 2 Qwen3.7-Plus ADS tools 27.8
  3. 3 Qwen3.7-Max ADS tools 27.6
  4. 4 Claude Fable 5 ADS tools 26.8
  5. 5 Claude Opus 4.8 Native search 25.2
  6. 6 GPT-5.6 Sol Native search 22.2
  7. 7 Gemini 3.5 Flash ADS tools 19.9

Source: Official SciTaste-Bench leaderboard

#ModelScoreSettingDateSource
1 Qwen 3.8 Max 28.0 ADS tools, DimAgent-v3 Aug 2026 Official leaderboard
2 Qwen3.7-Plus 27.8 ADS tools, DimAgent-v3 Jun 2026 Official leaderboard
3 Qwen3.7-Max 27.6 ADS tools, DimAgent-v3 May 2026 Official leaderboard
4 Claude Fable 5 26.8 ADS tools, DimAgent-v3 Jun 2026 Official leaderboard
5 Claude Opus 4.8 25.2 Native search, DimAgent-v3 May 2026 Official leaderboard
6 GPT-5.6 Sol 22.2 Native search, DimAgent-v3 Jul 2026 Official leaderboard
7 Gemini 3.5 Flash 19.9 ADS tools, DimAgent-v3 May 2026 Official leaderboard

Rankings by part

We read 9 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Chemistry, Medicine and Mathematics. Qwen3.7-Plus leads 5 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Chemistry

7 models we track have a score here.

  1. 1 Qwen 3.8 Max 27.8
  2. 2 Qwen3.7-Plus 26.8
  3. 3 Qwen3.7-Max 26.4
  4. 4 Claude Opus 4.8 ↑1 25.5
  5. 5 Claude Fable 5 ↓1 25.5
  6. 6 GPT-5.6 Sol 24.2
  7. 7 Gemini 3.5 Flash 19.7

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
Qwen 3.8 Max DimAgent-v3
Native sear…
28.0
Qwen3.7-Plus DimAgent-v3
Native sear…
27.8
Qwen3.7-Max DimAgent-v3
Native sear…
27.6
Claude Fable 5 DimAgent-v3
Native sear…
26.8
Claude Opus 4.8 DimAgent-v3
ADS tools
25.2
Claude Fable 5
Pure QA Native Sear…
24.6
Qwen 3.8 Max
Pure QA Native Sear…
23.2
GPT-5.6 Sol DimAgent-v3
ADS tools Native sear…
22.2
Claude Opus 4.8
Pure QA
21.9
GPT-5.6 Sol
Pure QA
21.1
Qwen3.7-Max
Pure QA
20.2
Gemini 3.5 Flash DimAgent-v3
Native sear…
19.9

Source: Official SciTaste-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen 3.8 Max ADS tools DimAgent-v3 28.0
Qwen3.7-Plus ADS tools DimAgent-v3 27.8
Qwen3.7-Plus Native search DimAgent-v3 27.7
Qwen3.7-Max ADS tools DimAgent-v3 27.6
Qwen3.7-Max Native search DimAgent-v3 27.4
Qwen 3.8 Max Native search DimAgent-v3 27.1
Claude Fable 5 ADS tools DimAgent-v3 26.8
Claude Opus 4.8 Native search DimAgent-v3 25.2
Claude Fable 5 Native search DimAgent-v3 25.2
Claude Fable 5 Native Search 24.6
Claude Opus 4.8 ADS tools DimAgent-v3 23.7
Qwen 3.8 Max Native Search 23.2
Claude Fable 5 Pure QA 22.3
GPT-5.6 Sol Native search DimAgent-v3 22.2
Claude Opus 4.8 Native Search 21.9
GPT-5.6 Sol Native Search 21.1
Qwen 3.8 Max Pure QA 21.1
Qwen3.7-Max Native Search 20.2
Claude Opus 4.8 Pure QA 20.2
Gemini 3.5 Flash ADS tools DimAgent-v3 19.9
Qwen3.7-Plus Native Search 19.8
Gemini 3.5 Flash Native search DimAgent-v3 19.7
GPT-5.6 Sol Pure QA 19.5
Qwen3.7-Plus Pure QA 19.4
Qwen3.7-Max Pure QA 19.3
Gemini 3.5 Flash Pure QA 16.9
Gemini 3.5 Flash Native Search 16.4
GPT-5.6 Sol ADS tools DimAgent-v3 9.50

Number one over time

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

Qwen
Own leaderboard Not tested since Oct 2026

Source: Official SciTaste-Bench 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
5.00 10.0 15.0 20.0 25.0 30.0 2026 5.00 10.0 15.0 20.0 25.0 30.0 2026

Source: Official SciTaste-Bench 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

All 9 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (7 entries).
  • Pass Models scored 7 models we track.
  • Pass Newest score The newest is from August 2026, for Qwen 3.8 Max.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Claude Fable 5 (#12).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 28 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.

How we check