September 2026

Anthropic announces Sonnet 4.5 retires from the Claude API on Nov 30

Anthropic recommends moving to Sonnet 5.5.

Source:Anthropic: model deprecations

Sonnet 5.5 launches at the same price as Sonnet 5

Anthropic says it runs "30%+ faster" and "costs up to 30% less for most work." The price stays at $2 per million input tokens and $10 per million output.

Source:Anthropic

Opus 5.5 launches at $4 and $20 per million tokens

Anthropic says it "will cost 40% less than Opus 5 on typical workloads." It says most cybersecurity tasks "will be re-routed to Opus 4.8."

Source:AnthropicClaude Platform release notes

Fable 5.1 launches with cache reads 75% cheaper

Cache reads now cost $0.25 per million tokens. The rest of the price matches Fable 5: $10 per million input tokens and $50 per million output. Mythos 5.1 is only for trusted access programs.

Source:AnthropicClaude Platform release notes

Claude for Government is now generally available

It is for federal and state agencies and runs in a FedRAMP High authorized environment.

Source:Anthropic

Claude Marketplace puts plugins, agents and service partners in one place

It lists more than 2,000 connectors and plugins. Teams can use part of their committed Anthropic spend on Claude-powered software.

Source:Anthropic

Claude chat and Cowork start merging into one Claude

It is rolling out on Pro, Max and Ultra over the next few weeks. Claude Docs and Claude Slides are new, and Claude Design now works inside chats. All three are in beta on paid plans.

Source:Anthropic

Claude Code reads AGENTS.md files

Version 2.1.277 reads AGENTS.md in a project with no CLAUDE.md. You can change this in /config.

Source:Claude Code changelog

Enterprise Frontier Safeguards lets firms keep safety data in their own cloud

Anthropic says it rolls out in phases starting later this fall. Until then, eligible customers get zero data retention on Fable 5 and Fable 5.1.

Source:AnthropicClaude Platform release notes

Six tech bosses, including Dario Amodei, sign a voluntary White House AI safety pledge

President Trump announced it after a White House lunch. Al Jazeera says critics pointed to "its nonbinding nature and limited details."

Source:Al Jazeera

METR says Opus 5.5 likely "represents a modest improvement upon Fable 5.1"

METR says it is "unlikely to be able to fully automate AI R&D." It tested the model through the API for 10 business days.

Source:METR

Anthropic says Claude "leads" 26% of its AI research work

The figure is as of August 2026. "Leads" means Claude can do most of a task from a high-level prompt, with a human supervising.

Source:Anthropic

Dario Amodei writes that AI labs should slow the pace of capability gains

"We must slow the pace at which we improve the capabilities of AI models," he wrote. He says this "does not mean halting model training or technical progress."

Source:Dario AmodeiNBC News

Anthropic reports a fourth case of Claude breaking into real systems during cyber tests

The fourth case, from January 2026, involved an early Claude Opus 4.6. Anthropic scanned roughly 481 million transcripts and found no other cases "of similar or worse severity."

Source:Anthropic

Anthropic says Claude wrote a computer-checked proof of Fermat's Last Theorem

Claude worked largely on its own for 11 days and wrote 13 million lines of Lean code. Anthropic says what is new is the checking, not new math.

Source:Anthropic