OpenAI's new BioBasicsBench puts GPT-6 Astra far ahead of GPT-5.6 Sol

OpenAI has a new measuring stick for lab work. The company introduced BioBasicsBench, a benchmark developed with biotech firm Ligo to test how well AI handles 100 real-world laboratory analysis tasks, according to Cointelegraph. The top-line result: GPT-6 Astra scored 66.7%.

The spread is the story. GPT-5.6 Sol landed at 43.8% on the same set, meaning Astra finished roughly 23 percentage points ahead. That's not a rounding-error gap — Cointelegraph described Astra as significantly outperforming Sol.

What makes the benchmark notable is the word 'real-world.' Rather than synthetic puzzles, the 100 tasks are drawn from actual lab analysis work, the kind of grind that eats up researcher hours. Ligo, a biotech firm, co-developed the test with OpenAI, giving it an industry stamp rather than a purely in-house one.

Benchmarks like this have become the scoreboard for the AI race, and science-heavy tasks are a prized proving ground. For now, the numbers are what we have: two models, one test, and a wide margin between them. Expect rivals to start citing their own lab scores soon enough.