Fable 5 returns, Claude Sonnet 5 launches, Claude Science ships — all on the same day

Anthropic released Claude Sonnet 5 on June 30 as the default model for all Free and Pro users — and for the first time, a Sonnet-class model can plan, use tools, browse, and run autonomously at a level that previously required Opus. It scores 63.2% on SWE-Bench Pro (vs 58.1% for Sonnet 4.6 and 69.2% for Opus 4.8), slightly outperforms Opus 4.8 on knowledge work benchmarks, ships with a 1M token context window and 128K max output, and has adaptive thinking on by default. Introductory API pricing is $2 per million input tokens and $10 per million output tokens through August 31, 2026, then rises to standard pricing of $3/$15 — the same sticker as Sonnet 4.6. Available via claude-sonnet-5 on the API, Claude Code, Amazon Bedrock, and Microsoft Foundry; Google Vertex AI coming soon.

Sonnet 5 is a drop-in upgrade for Sonnet 4.6 with three behavior changes that will break existing integrations: (1) Adaptive thinking is on by default — requests without a thinking field now run with thinking enabled; pass thinking: {type: 'disabled'} to opt out, and revisit any max_tokens cap that assumed no thinking tokens. (2) Setting temperature, top_p, or top_k to a non-default value now returns a 400 error — remove these parameters before migrating. (3) Manual extended thinking (thinking: {type: 'enabled', budget_tokens: N}) is removed entirely and returns a 400 error. Also note: the new tokenizer produces approximately 30% more tokens for the same text, which affects any token budget or cost estimate derived from Sonnet 4.6 counts.

The US Commerce Department lifted its export control directive on June 30, and Anthropic restored Fable 5 globally on July 1 across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. The restoration comes with a new safety classifier that blocks the reported jailbreak technique in over 99% of cases — blocked requests fall back to Opus 4.8 with a stated reason. The trigger was an Amazon researcher finding a prompt that got the model to flag software flaws and write proof-of-concept exploit code. Tom Brown (Anthropic co-founder) led the final negotiations with the Trump administration after replacing Dario Amodei in that role. Fable 5 is included in Pro, Max, Team, and select Enterprise plans at up to 50% of weekly usage limits through July 7, after which usage credits are required.

Claude Science is a new platform that integrates over 60 scientific databases and computation tools — protein structure prediction, molecular modeling, literature search, lab data analysis — in a single workspace. It produces auditable artifacts, supports flexible compute allocation, and is designed for pharmaceutical research teams, academic biology labs, and biotech startups. Dario Amodei presented at a San Francisco launch event, calling it Anthropic's answer to what Claude Code did for software engineering. Claude Science is Anthropic's direct competitor to OpenAI's GPT-Rosalind (launched April 2026). Early access is available at anthropic.com/science.

As part of the Fable 5 restoration announcement, Anthropic proposed a shared industry standard for scoring how dangerous a jailbreak really is — co-developed with Amazon, Microsoft, Google, and other Glasswing partners. Each jailbreak gets scored on four dimensions: capability gain (how much further it takes a user beyond existing tools), breadth (how many distinct attacks it unlocks), ease of weaponization (skill and effort required), and discoverability (how easy the technique is to find or replicate). For the highest-severity cases, Anthropic commits to deploying fixes the moment severity is confirmed. A 24/7 jailbreak monitoring team is being stood up, and a HackerOne bug bounty program is now open for Fable 5 jailbreak reports.

Anthropic published a detailed post-mortem on the 19-day export control ban. Amazon researchers discovered the jailbreak. Dario Amodei was replaced by co-founder Tom Brown in Commerce Department negotiations — reportedly because Amodei was a target of the Trump administration given his vocal support for Kamala Harris in 2024. Commerce Secretary Lutnick's restoration letter was addressed to Brown, not Amodei. The incident occurred under a June 2 executive order that created a voluntary path for frontier model review before release, with a classified benchmark to determine which models are 'covered.' Anthropic agreed to coordinate on future launches and report any malicious use detected.

Claude Code changed the default permission mode from 'default' to 'Manual' across the CLI, VS Code, and JetBrains. This means Claude will pause and ask for confirmation before taking actions rather than proceeding automatically — a more conservative default for new users and agentic workflows. Separately, AskUserQuestion dialogs no longer auto-continue by default; opt into an idle timeout via /config. Also in this release: Claude Sonnet 5 sessions no longer use the mid-conversation system role for harness reminders, fixing a session behavior bug introduced at Sonnet 5 launch.