Today we’re announcing that we’ve activated ASL-3 protections for Claude Opus 4—our most stringent AI safety measures yet. This is the culmination of two years of focused effort to race to the top on responsible scaling. 🧵

@DanHendrycks I would like to point out that this kind of amnesia (which I prefer to call "explicit memory") is advantageous from an AI safety perspective. 1/2

Claude can now use Skills. Skills are packaged instructions that teach Claude your way of working.

Introducing Claude Haiku 4.5: our latest small model. Five months ago, Claude Sonnet 4 was state-of-the-art. Today, Haiku 4.5 matches its coding performance at one-third the cost and more than twice the speed.

Technological Optimism and Appropriate Fear - an essay where I grapple with how I feel about the continued steady march towards powerful AI systems. The world will bend around AI akin to how a black hole pulls and bends everything around itself.

Sonnet 4.5 Tops LiveBench For Agentic Coding The best models that can actually build things are the ones that excel at agentic code. Sonnet 4.5 tops the leaderboard with GPT-5 hot on its heels. Sonnet 4.5 (thinking) also scores #1 on the non-agentic coding category! Making it

We ran one of our hardest computer-use benchmarks on Anthropic Sonnet 4.5, side-by-side with Sonnet 4 ↴↴ 1/4

Sonnet 4.5 tops both writing evals! On spiral-bench: much stronger pushback & de-escalation vs sonnet-4. GLM-4.6's score is incremental over GLM-4.5 - but personally I like the newer version's writing much better. Links & writing samples ->

Sonnet 4.5 crushing GPT-5 high on ARC-AGI 2

massive jump in medium and hard cybersecurity CTF challenges with Sonnet 4.5

Prior to the release of Claude Sonnet 4.5, we conducted a white-box audit of the model, applying interpretability techniques to “read the model’s mind” in order to validate its reliability and alignment. This was the first such audit on a frontier LLM, to our knowledge. (1/15)

AI agents are now capable of doing real, if bounded, work. But that work can be very valuable. For example, the new Claude Sonnet 4.5 was able to replicate published economics research from data files & the paper. We need to figure out what to do with it:

Sonnet 4.5, meet Droid. After joint testing with @AnthropicAI, we find the strengths of Sonnet 4.5 to be: • Significantly more reliable and accurate file editing • High environmental awareness • Snappier than previous models on quick questions, not overthinking simple asks

Wrote up my initial impressions of the brand new Claude Sonnet 4.5 - I think it may live up to Anthropic's claims of being the "best coding model in the world", for the next few weeks at least! https://t.co/NQlytA1Xlm

Huge amount of effort went into augmenting 4.5 for cyber use with our partners. We think that 4.5 is the best model for cyber defense now.

https://t.co/iIHe4HWVv4 Not gonna lie: Claude Chrome extension looks like the CUA we were waiting for! Lets go!

Claude Sonnet 4.5 is here and I got to try it for about 2 days. I would say It's the strongest model for building complex agents. Multi-agents system worked for the first time for me . Let's see what crazy agents people come up with.

Claude Sonnet 4.5 is now the default model for Augment Code. We’re rolling it out to all customers over the next 24 hours, where it will be available alongside Sonnet 4 (for a limited time) and GPT-5 in the model picker. Here’s how Sonnet 4.5 compares to Sonnet 4:

is it just me or has claude 4.5 become dumber for everyone in the past few hours?

BREAKING: Anthropic just dropped Claude Sonnet 4.5! We've been testing it for a few days @every and here's what we found: - It's smarter and faster than Opus: It solved a nasty bug for @kieranklaassen than Opus 4.1 was continually failing at. And it feels twice as fast. - It's

Shipping today: @AnthropicAI's Sonnet 4.5 in Notion. Better reasoning, smarter planning, and improved understanding that makes your personal Agent feel truly personalized. Built to get work done.