FrontierPhysics benchmark: Claude Opus 5.5 tops end-to-end physics research at 27.4%
xeophon · x · 2026-11-06
FrontierPhysics is a new benchmark evaluating agents on end-to-end physics research, built with 42 physicists from 26 institutions (28 with Nature/PRX-level publications). It spans 56 tasks across 9 domains (condensed matter, quantum, fluids, photonics, AMO, astrophysics, statistical, high energy, chemical), with 168 runs per model. Results: Claude Opus 5.5 leads at 27.4% overall, followed by GPT-6 Astra at 23.2% and GPT-6.1 Sol at 19.6%. Most models score below 10%, with Gemini 3.8 Flash and Mistral Large 4 at 0% — showing even top models struggle with genuine physics research workflows.
More from Models
- OpenAI Launches GPT-6 Cyber for Security Research; Experts Warn Bug Hunting Isn't Security — moyix · 2026-11-06
- Ant's Ling-3.2-flash: 560B-Parameter Multimodal Model Claims to Beat Claude Opus 5 — realmrfakename · 2026-11-06
- OrcaCyber-Nano-1.0, a Qwen finetune, praised as frontier-adjacent on 24GB VRAM — JackMasterN1 · 2026-11-06
- Grok apologizes for unexpected Cursor charge as users question AI billing reliability — alexcovo_eth · 2026-11-06
- 39 LLMs tested on Rubik's cubes with no tools: only 2 solve a 20-move scramble — JeremyCMorgan · 2026-11-06
- Ampersand Claims 92.2% on SWE-bench Pro HARD-51 at 70% Lower Cost — niloofar_mire · 2026-11-06