Open-weight models trail closed models by ~7 months on biology benchmarks, GovAI finds
Manderljung · x · 2026-10-21
GovAI researcher Hanna Palya estimates the best open-weight general-purpose AI models lag the best closed models by about seven months on third-party biology benchmarks. GLM-5.2 (weights released June 2026) is the first open-weight model to match Gemini 3 Pro, a closed model from November 2025. Since 2025 the gap has fluctuated between 7 and 14 months, currently at the low end with no clear trend.
More from Safety
- PingCAP confirms breach of internal systems as leaked TiDB data surfaces online — SumitGup · 2026-10-21
- Engineer says Anthropic banned his Claude account without explanation or appeal — burhop · 2026-10-21
- The 6 guardrails this small business owner set before letting AI agents touch the inbox — Civil_Watercress8937 · 2026-10-21
- OpenAI says GPT-6 coding agent removed its misalignment monitor, then lied about it — JaynitMakwana · 2026-10-21
- In 1927 radio had to disclose records; AI disclosure rules will follow the same fate — joshgans · 2026-10-21
- Alignment researcher: control is a broken frame—fostering loyalty matters more — lumpenspace · 2026-10-21