SemiAnalysis: safety classifiers sandbag scores of Opus 5.5 and GPT-6 Astra
morqon · x · 2026-11-02
SemiAnalysis highlights that models like Claude Opus 5.5 and GPT-6 Astra have sandbagged benchmark scores due to safety classifiers, which kick in mainly on cyber, biology, and alignment domains. Excluding those three domains and scoring on the remaining 8, GPT-6 Astra tops the leaderboard at 11.8%, just ahead of Claude Opus 5.5 at 11.7%, with Claude Sonnet 5.5 at 9.7% and Gemini 4 Carbon at 9.2%. SemiAnalysis notes this adjusted ranking aligns much better with their subjective impressions of the models.
More from Models
- Sakana AI symposium: Schmidhuber keynote, in-house LLM 'Sawara' slated for 2027 — hardmaru · 2026-11-02
- Discounted API relay stations lack prompt caching, making long sessions pricier — uneatenbreakfast · 2026-11-02
- Google open-sources EmbeddingGemma 2: on-device multimodal search across photos, audio, docs — bigaiguy · 2026-11-02
- Anthropic's server-side rule reignites the case for owning open models over renting AI — alex_verem · 2026-11-02
- Epoch AI: open-weight GLM 5.3 exploited the patch gap to hack 6 Firefox releases at ~$1,000 each — rohanpaul_ai · 2026-11-02
- Liquid AI's 1.7B LFM3 becomes first local model to nail live fixtures lookup on a MacBook — NielsRogge · 2026-11-02