Benchmark Almanac

BioSimBench

CellType 27B 0.26 tops its leaderboard Gemini 4 Argon 0.18 leads the models we track
Not in model cards

Models predict which genes change after drugs, gene knockdowns, and immune signals across four prediction tracks containing 60, 60, 60, and 27 cases. The overall score averages perturbation-specific F1 across those tracks on a zero-to-one scale.

CellType keeps test answers private and offers hosted evaluations. Models receive separate scores on 296 research-agent episodes, which do not enter the prediction average. CellType terms prohibit automated access and systematic data collection without permission.

Made by
CellType
Released
November 2026
Kind
Genomics and life sciences
Level
Expert
Scored
Checked automatically
Answers
Short answer
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
14
Entries on its own leaderboard, 12 models we track

CellType 27B tops the leaderboard with 0.26

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. 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 Gemini 4 Argon, at 0.18.

14 entries. Scroll sideways for the rest.

  1. 0.26 CellType 27B
  2. 0.19 CellType 9B
  3. 0.18 Gemini 4 Argon
  4. 0.16 Claude Sonnet 5.5
  5. 0.16 GPT-6 Astra
  6. 0.15 Gemini 3.1 Pro
  7. 0.15 GPT-6.1 Sol
  8. 0.15 Gemini 3.8 Flash
  9. 0.15 Claude Opus 5.5
  10. 0.14 GPT-6 Sol
  11. 0.13 Claude Haiku 5.5
  12. 0.12 GPT-6 Luna
  13. 0.12 Grok 4.7
  14. 0.11 DeepSeek-V4-Pro
  1. 1 CellType 27B trained in BioSimGym, same prompt as the frontier models, answer = vote over 9 answers (1 plus 8 samples) 0.26
  2. 2 CellType 9B trained in BioSimGym, same prompt as the frontier models 0.19
  3. 3 Gemini 4 Argon high reasoning effort (highest) 0.18
  4. 4 Claude Sonnet 5.5 max reasoning effort 0.16
  5. 5 GPT-6 Astra max reasoning effort 0.16
  6. 6 Gemini 3.1 Pro high reasoning effort (highest) 0.15
  7. 7 GPT-6.1 Sol max reasoning effort 0.15
  8. 8 Gemini 3.8 Flash high reasoning effort (highest) 0.15
  9. 9 Claude Opus 5.5 max reasoning effort 0.15
  10. 10 GPT-6 Sol max reasoning effort 0.14
  11. 11 Claude Haiku 5.5 max reasoning effort 0.13
  12. 12 GPT-6 Luna max reasoning effort 0.12
  13. 13 Grok 4.7 high reasoning effort 0.12
  14. 14 DeepSeek-V4-Pro max reasoning effort 0.11

Source: Official BioSimBench leaderboard

#EntryScoreSettingDateSource
1 CellType 27B 0.26 trained in BioSimGym, same prompt as the frontier models, answer = vote over 9 answers (1 plus 8 samples) Official leaderboard
2 CellType 9B 0.19 trained in BioSimGym, same prompt as the frontier models Official leaderboard
3 Gemini 4 Argon 0.18 high reasoning effort (highest) Sep 2026 Official leaderboard
4 Claude Sonnet 5.5 0.16 max reasoning effort Sep 2026 Official leaderboard
5 GPT-6 Astra 0.16 max reasoning effort Sep 2026 Official leaderboard
6 Gemini 3.1 Pro 0.15 high reasoning effort (highest) Feb 2026 Official leaderboard
7 GPT-6.1 Sol 0.15 max reasoning effort Sep 2026 Official leaderboard
8 Gemini 3.8 Flash 0.15 high reasoning effort (highest) Sep 2026 Official leaderboard
9 Claude Opus 5.5 0.15 max reasoning effort Sep 2026 Official leaderboard
10 GPT-6 Sol 0.14 max reasoning effort Sep 2026 Official leaderboard
11 Claude Haiku 5.5 0.13 max reasoning effort Oct 2026 Official leaderboard
12 GPT-6 Luna 0.12 max reasoning effort Sep 2026 Official leaderboard
13 Grok 4.7 0.12 high reasoning effort Sep 2026 Official leaderboard
14 DeepSeek-V4-Pro 0.11 max reasoning effort Apr 2026 Official leaderboard

Number one over time

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

Google DeepMind
Own leaderboard 1 leader
Gemini 4 Argon 0.18

Source: Official BioSimBench leaderboard

Scores over time

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

On the benchmark’s own leaderboard
0.10 0.13 0.15 0.17 0.20 2026 0.10 0.13 0.15 0.17 0.20 2026

Source: Official BioSimBench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Biology and chemistry

All 108 in Biology and chemistry
  • GeneBench Pro In use Agents analyze messy simulated genomics and biology datasets to estimate a decision-relevant quantity across 129 problems. Scored automatically against known answers. GPT-5.6 Sol 31.5%
  • SUPERChem In use Models solve 500 expert-written chemistry reasoning problems, offered in multimodal and text-only forms. Expert solution paths also allow a reasoning path score. Tencent Hy4 preview 66.4%
  • LifeSciBench In use Models complete 750 expert-written life science research tasks, often with attached figures, tables, or sequence files. Graded against expert rubrics as score and pass rate. GPT-Rosalind-6 61.3%
  • MedChemBench (OpenAI) In use Models answer chemistry questions from an OpenAI internal set, grouped with OpenAI's life science research evaluations. OpenAI has not published the task format. GPT-Rosalind-6 49.6%
  • ProtocolQA Open-Ended In use Models correct errors in published lab protocols using open-ended questions adapted from LAB-Bench ProtocolQA. Scored as pass@1. GPT-6 Cyber 47.2%
  • MolecularIQ In use Models answer questions about molecular graphs, such as counting, indexing and constrained generation, and a symbolic checker verifies each answer. gpt-oss-120b 47.5%