Model WarsJune 18, 2026via MarkTechPost
OpenAI Releases LifeSciBench, a 750-Task Benchmark Grading AI Models on Real Life-Science Research With Expert-Written Rubric
Why it matters
OpenAI's LifeSciBench reveals a critical capability gap: frontier models struggle with domain-specific, reasoning-heavy tasks in high-stakes research. This matters because it shows where AI adoption in biotech will stall—and where the next wave of model improvement must focus.
Key signals
- 750-task benchmark across 7 biological domains and 7 research workflows
- Built by 173 PhD scientists with 19,020 rubric criteria
- GPT-Rosalind (best model) passes only 36.1% of tasks
- Evaluation grades reasoning and decision-making, not just recall
- Significant headroom identified in artifacts, exact outputs, and operational calls
The hook
36.1%. That's how many real life-science research tasks the best AI model passes. OpenAI just released LifeSciBench—750 expert-authored benchmarks that grade reasoning, not recall.
OpenAI's LifeSciBench evaluates whether frontier AI can handle real life-science research across 750 expert-authored tasks, seven workflows, and seven biological domains. Built by 173 PhD scientists with 19,020 rubric criteria, it grades reasoning and decisions, not just recall. The best model, GPT-Rosalind, passes 36.1%, leaving large headroom on artifacts, exact outputs, and operational calls.
The post OpenAI Releases LifeSciBench, a 750-Task Benchmark Grading AI Models on Real Life-Science Research With Expert-Written Rubric appeared first on MarkTechPost.