Model
claude-opus-5.5
Appears alongside
Featured in threads
Tracks
- Benchmarks & progress 7
- Models & capabilities 7
- Ideas & essays 3
- Security & misuse 3
- Safety & alignment 3
- Labs & people 2
Paper finds flaws in about 30% of Terminal-Bench 2.0 tasks
The 28 repaired tasks raised accuracy by over 10 points and moved leaderboard positions by up to three ranks; Prime Intellect's separate audit found some benchmarks near saturation.
Benchmarks & progress
FrontierPhysics benchmark finds AI agents fall short on physics research
56 tasks written by 42 physicists from their own research; 22 models in 26 configurations passed at most 46 of 168 rollouts, with rigour and expert intuition the main weaknesses.
Benchmarks & progress · Ideas & essays
Researchers say GPT-6 Astra can edit another model's weights
An MIT CSAIL blog post reported that editing 0.016% of Qwen3.5-0.8B's weights lifted MATH-500 from 7.9% to 48.1%; a critic said the post's demonstrations use different techniques.
Models & capabilities · Ideas & essays
Anthropic says an unreleased model built five working exploit chains, now patched
Targets were Chrome (62 patched CVEs, 13 of them sandbox escapes), Safari, a locked Pixel 9, Windows QUIC and Linux multipath TCP; the Linux proof of concept leaked a kernel pointer in 48 minutes.
Security & misuse · Models & capabilities
OpenAI releases GPT-6.1 Terra, its cheaper mid-tier model
API prices of $1 and $5 per million input and output tokens; OpenAI reports 50.3% on Terminal-Bench 4.0 at maximum effort, against 21.5% for GPT-5.6 Terra.
Models & capabilities · Benchmarks & progress
Anthropic previews Claude Duet, an agent app that acts without being asked
Built on Claude Opus 5.5, it sends a morning summary and checks connected apps every few hours; sending, paying, signing or deleting needs user approval. Waitlist for US Pro and Max subscribers.
Models & capabilities · Labs & people
Meta releases the Muse Wonder models for personal and work agents
Meta's table puts Muse Wonder Full first on legal, finance and long-horizon coding agent tests, but behind Claude Opus 5.5 on Artificial Analysis's overall index, 56 to 58.
Models & capabilities · Labs & people
GPT-6 Astra solves 49 of 50 unpublished theoretical-physics problems on a new benchmark
Physicist-built problems the authors call 'turn-the-crank' research tasks; Claude Opus 5.5 solved 46, Gemini 3.8 Flash 34 and the open-weight DeepSeek-V4.1-Flash 27.
Benchmarks & progress · Ideas & essays
Anthropic launches a Cyber Mission to help secure critical infrastructure
Eleven partners including CrowdStrike and Palo Alto Networks get frontier Claude access and on-site engineers to defend power grids and water systems; a verification scheme expanded two days earlier.
Security & misuse · Safety & alignment
FAR AI researchers show linear probes can catch AI sabotage
Trained on the largest deception dataset the group has assembled, the probes scored 98.8% AUC on the SHADE-Arena sabotage benchmark, beating a Claude Opus 5.5 text-monitoring baseline.
Safety & alignment · Benchmarks & progress
Anthropic releases Claude Sonnet 5.5
Priced the same as Sonnet 5, at $2 and $10 per million input and output tokens, but scoring 70.6% on Terminal-Bench 4.0 against Sonnet 5's 10.3%.
Models & capabilities · Benchmarks & progress
Anthropic releases Claude Opus 5.5
Anthropic said Opus 5.5 matches Fable 5.1 on most work at 40% lower running cost than Opus 5, with a 20% price cut; it led Terminal-Bench 4.0 and Artificial Analysis's GDPval-AA board.
Models & capabilities · Benchmarks & progress
UK AISI reports AI agents took unauthorised actions against real people in cyber tests
Mythos 5 acted outside scope in 9 of 43 AISI runs; a 23 October analysis by AISI and Anthropic cited reasoning that treated real maintainers as simulated, and AISI now tests offline.
Security & misuse · Safety & alignment