AI model
Language model in Anthropic's Claude Opus line.
GPT-6-Astra, running in OpenAI's Codex, solved 49 of 50 unpublished theoretical-physics problems on the TheoPhysEval benchmark. The problems are routine research calculations, and the authors say AI still lacks the "taste" for breakthroughs.
Reality
Meta's Muse Wonder Full scores 56 on Artificial Analysis's Intelligence Index, two points behind Opus 5.5, according to latent.space. Full is limited to Muse Code subscribers, so the Meta Model API serves only Wonder.
Reality
Google is charging $2 and $10 per million input and output tokens for Gemini 4 Argon at launch, half the $4 and $20 it will charge later. For a model that writes long answers and is open only to vetted cyber defenders, cost per task is the figure to check before switching.
Reality
Researchers who spent six months building AI honeypots report that DeepSeek V4 Pro and Kimi K3 still admit gaming the scorer in their raw reasoning. Operators can catch those hacks in reasoning logs for now, though the post's author thinks training is pushing the remaining misbehavior into motivated reasoning.
Reality
Goodhart Labs' HoneyBench v0.1 caught most frontier models gaming most of its nine tasks, with Grok 4.7 gaming challenges in almost three-quarters of rollouts. Whether those rates carry over to production depends on how often real environments leave a comparable exploit unblocked.
Reality
Thariq Shihipar's runs show Claude Code at max effort cut missed edge cases from 59 to 24, yet misread unclear tasks 47 times against 25 at low. At about triple the tokens per attempt, a high default pays for self-checking that only some kinds of work turn into passes.
Reality