Breaking

OpenAI Unveils BioBasicsBench, Where GPT-6 Astra Scores 66.7% on Lab Analysis Tasks

November 6, 2026 at 20:07 EST

Research & Papers
Healthcare

OpenAI has introduced BioBasicsBench, a new benchmark developed with biotech firm Ligo to measure how well AI models perform on 100 real-world laboratory analysis tasks, with its new GPT-6 Astra model scoring 66.7% against 43.8% for GPT-5.6 Sol.

OpenAI · Ligo

A new lab-task benchmark shows GPT-6 Astra ahead of GPT-5.6 Sol by 22.9 points

OpenAI, working with biotech company Ligo, introduced BioBasicsBench: 100 tasks drawn from real laboratory analysis work. The developer reports 66.7% accuracy for GPT-6 Astra and 43.8% for GPT-5.6 Sol.

66.7%

GPT-6 Astra accuracy

43.8%

GPT-5.6 Sol accuracy

+22.9

percentage-point gap

100

real-world lab tasks in the benchmark

Reported accuracy on BioBasicsBench

Column height is proportional to accuracy on a 0–100% scale. Each stacked band marks 10 points.

66.7%

GPT-6 Astra

43.8%

GPT-5.6 Sol

The gap between the two columns is about 23 points, from a single reported benchmark of 100 tasks.

Why it matters

Targets practical lab analysis rather than general knowledge or abstract reasoning.
Relevant to drug discovery, diagnostics and research throughput.
Built with an industry partner, grounding the tasks in real scientific workflows.

What is still unclear

Methodology, grading criteria and task categories are not detailed.
No independent third-party verification is available yet.
Public release, and whether other developers will be tested, is not confirmed.

Read the headline carefully

These are the developer's reported results, pending broader scrutiny.

Single benchmark 100 tasks Methodology not yet published Independent testing pending

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.