Astra's Hyra 2.0 trains protein models with SOTA scores and 57× speedup
zdhnarsil · x · 2026-11-11
Astra announced two protein models trained from scratch with its Hyra 2.0 system (an 'AI4AI4Science' approach where AI builds SOTA models):
- HyInv: a protein inverse-folding model with a SOTA scTM score and 57× the speed of ProteinMPNN (2,532 designs/s), pushing the accuracy-speed Pareto frontier well past prior methods.
- HyPLM-Flash: a 650M protein language model matching or beating ESM-2 and ESM C at comparable scale, with an embedding throughput of 131k residues/s — the fastest in its class.
The team argues weight-space programming alone won't scale AI R&D, and positions AI-built models as the next era for AI-driven science.
More from Models
- OpenAI to retire Advanced and Standard voice in ChatGPT on Dec 11, going all-in on GPT-Live — OpenAINewsroom · 2026-11-11
- TheZi slams Anthropic's "honeypot" eval as Claude Mythos 5.2 transparently answers 84 over truer 88 — TheZvi · 2026-11-11
- Gemini 4 Carbon wins across runs in a sailing sim race, beating Fable and Astra — shekitup · 2026-11-11
- X's recommender is multimodal: CLIP and MultimodalPostEmbedderV5 embeddings feed Phoenix retrieval — cocktailpeanut · 2026-11-11
- Anthropic grants paid users a free full usage reset before Dec 10 after limit bugs — ClaudeDevs · 2026-11-11
- Gemini 4 Argon safety report appears to describe CoT training, contradicting GDM's claims — S_OhEigeartaigh · 2026-11-11