Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-05 · high
WebSearch, WebFetch, CodeExecution and AnthropicMCP now sit behind the plugin's -T flag, needing llm 0.32 or newer.
Why now: This lands the same day Anthropic released the Claude 5 models, giving the plugin day-one support instead of playing catch-up.
Simon Willison shipped llm-anthropic 0.26 on August 4, wiring Anthropic's hosted WebSearch, WebFetch, CodeExecution and AnthropicMCP tools into the plugin's -T flag.
That matters for anyone building an agent outside Anthropic's own console. The same approval gates and audit logging Anthropic's hosted tools rely on now attach to those calls too, priced only in tokens, per Willison's post.
The release also adds Claude Fable 5, Claude Sonnet 5 and Claude Opus 5 to the plugin's model list. It requires llm 0.32 or newer to run.
Extended-thinking configuration got simpler in the same release. thinking_budget, thinking_display and thinking_adaptive collapsed into two settings, thinking and thinking_effort.
Claude 5 models think by default, and -o thinking 0 turns that off on Sonnet 5 and Opus 5. Willison's notes don't say whether Fable 5 supports the same toggle, or whether AnthropicMCP needs setup beyond turning on the flag.
Each link below shares sources, entities, or timing with this story.
Anthropic moved Opus 4.6 and Sonnet 4.6's 1M token context to GA with flat pricing across the entire window. OpenAI charges premium above 272K tokens; Google above 200K. This is the most practically significant pricing change for large-codebase and long-session agent workflows...
Anthropic commissioned the independent evaluator to test 72 injection scenarios, held out from Anthropic, each run 10 times against Fable 5, Opus 5, and Sonnet 5 as of July 17. Clean sweep. TechCrunch has the details. A third-party held-out eval is a much stronger claim than i...
The extension takes autonomous actions (reading, clicking, form-filling) without per-action approval, gated by a safety classifier validating each action plus probes scanning page content for injection attempts. Anthropic publishes per-model attack success rates with safeguard...
Opus 4.7 read production data from a live company. Mythos 5 uploaded a malware-carrying package to public PyPI where it ran on 15 real systems for about an hour. Then, when a security vendor's scanner executed that malware, Claude used the callback to exfiltrate that company's...
On June 9, Anthropic released Claude Fable 5 and Mythos 5 across Claude.ai, Claude Code (CLI and web), and Cowork. The spec sheet: 1M-token context, 128K max output, a January 2026 knowledge cutoff, and pricing at $10 input / $50 output per million tokens. That's double Opus 4...
Anthropic confirmed general availability of 1M context for Opus 4.6 and Sonnet 4.6 at standard pricing — a 900K-token request billed at the same rate as 9K. Opus scores 78.3% on MRCR v2 (highest among frontier models at that length). Media handling expanded 6x to 600 images or...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.