Synthesis history

4 versions, newest first.

Version 4 2026-07-04 18:34 UTC · 134 items

The GeneBench-Pro methodology paper appeared on bioRxiv (v2) [^39706], the first formal academic record of the benchmark's design, though it is OpenAI's own work rather than independent review. Pause IA entered as a new…
Version 3 2026-07-03 08:37 UTC · 115 items

No substantive new claims, voices, or events this pass. All new items are social amplification of the original announcement — English, Spanish, and Japanese accounts sharing the OpenAI post, plus brief tech-news coverag…
Version 2 2026-07-02 02:41 UTC · 103 items

Two substantive additions this pass: @ollobrains offered a new framing that GeneBench-Pro tests whether AI can behave like a computational biologist rather than merely whether it knows biology [^38008], and Collin Burdi…
Version 1 2026-06-30 18:26 UTC · 72 items

OpenAI released GeneBench-Pro on June 30, 2026 — a 129-problem expert computational biology benchmark using synthetically generated problems with deterministic grading [^36672]. GPT-5.6 Sol scores 31.5% with Pro mode en…