BioBasicsBench: everyday wet-lab science (BioBasicsBench)
We show this table for reference; we do not rank on it.
OpenAI and Ligo's benchmark of how well AI agents interpret raw wet-lab evidence, such as gel and blot images, colony plates, microscopy, sequencing files and instrument exports, spot experimental errors and recognize when the data cannot support a conclusion.
Benchmark score on BioBasicsBench — November 6, 2026
We compile the BioBasicsBench rows from provider self-reports. GPT-6 Astra leads the table at 67.0%, followed by GPT-6.1 Sol (62.6%) and GPT-5.6 Sol (45.6%). We do not use these results to rank models overall.
GPT-6 Astra
OpenAI
GPT-6.1 Sol
OpenAI
GPT-5.6 Sol
OpenAI
7 modelsReasoningCurrentDisplay onlyUpdated November 6, 2026
Benchmark score table (7 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 67.0% | Not reported | Closed |
| 2 | GPT-6.1 SolOpenAI | 62.6% | Not reported | Closed |
| 3 | GPT-5.6 SolOpenAI | 45.6% | Not reported | Closed |
| 4 | GPT-6 LunaOpenAI | 37.6% | Not reported | Closed |
| 5 | GPT-5.5OpenAI | 35.4% | Not reported | Closed |
| 6 | GPT-5.4OpenAI | 33.2% | Not reported | Closed |
| 7 | GPT-5.2OpenAI | 31.2% | Not reported | Closed |
Among the reported BioBasicsBench rows, GPT-6 Astra is first at 67.0%. The third row is 21.4 points behind. The broader top-10 range is 35.8 points, so the table still separates the published systems.
7 models have been evaluated on BioBasicsBench. The benchmark falls in the Reasoning category. BioBasicsBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About BioBasicsBench
Year
2026
Tasks
100 routine wet-lab analysis tasks (88 multimodal) across five domains
Format
Mean reward (% of rubric credit), five attempts per task
Difficulty
Routine wet-lab data interpretation
Ligo scientists wrote the 100 tasks from their own experiments across protein methods, imaging and cytometry, quantification and assays, nucleic acid methods and molecular cloning; each task has an expert-written rubric and went through two internal reviews and one external PhD-level review. Agents return a structured answer and earn partial credit per rubric criterion; the score is mean reward (percent of available credit) over five attempts per task, with 95% whole-task bootstrap intervals. OpenAI ran every published result and has not released the tasks, so BenchLM shows the scores display-only and keeps them out of weighted rankings. BenchLM stores each model's highest published reasoning effort: max for GPT-6 Astra, GPT-6.1 Sol, GPT-5.6 Sol and GPT-6 Luna, xhigh for GPT-5.2, GPT-5.4 and GPT-5.5.
Freshness and provenance
Version
BioBasicsBench (100 tasks)
Refresh cadence
Static
Staleness state
Current
Question availability
Private provider task set with published aggregate results
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does BioBasicsBench measure?
OpenAI and Ligo's benchmark of how well AI agents interpret raw wet-lab evidence, such as gel and blot images, colony plates, microscopy, sequencing files and instrument exports, spot experimental errors and recognize when the data cannot support a conclusion.
Which model scores highest on BioBasicsBench?
GPT-6 Astra by OpenAI currently leads with a score of 67.0% on BioBasicsBench.
How many models are evaluated on BioBasicsBench?
7 AI models have published results on BioBasicsBench in the BenchLM catalog.
Compare top models on BioBasicsBench
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.