Claude Sonnet 5 launches: new model picker with Fable 5, Opus 4.8, Sonnet 5, and Haiku 4.5

Anthropic has released Claude Sonnet 5, and with it a refreshed model picker across Claude.ai, Claude Cowork, Claude Code, and the desktop apps. The lineup you now see when you open the model menu: Fable 5 ('For your toughest challenges'), Opus 4.8 ('For complex tasks'), Sonnet 5 ('Most efficient for everyday tasks'), and Haiku 4.5 ('Fastest for quick answers'). Older models such as Sonnet 4.6 remain available under 'More models'.

Sonnet 5 is the big news. It replaces Sonnet 4.6 as the recommended model for everyday work and closes much of the gap with the Opus tier: on coding and agentic tasks it reaches what was previously Opus-level quality, at Sonnet-tier speed and cost. It keeps the 1M-token context window, raises the output ceiling to 128K tokens, and has adaptive thinking on by default—the model decides for itself when and how deeply to reason.

Pricing: Sonnet 5 lists at $3 input / $15 output per million tokens, the familiar Sonnet price point. As a launch offer, Anthropic is running introductory API pricing of $2 input / $10 output per million tokens through August 31, 2026. For subscribers on Claude.ai and Cowork nothing changes: Sonnet 5 is simply included in your plan.

Fable 5: included until July 12. When Fable 5 returned on July 1, the free window (up to 50% of your weekly usage limit on Pro, Max, Team, and select Enterprise plans) was set to end July 7. Anthropic has extended it: the picker now shows 'Included until July 12'. After that date, additional Fable 5 usage runs on usage credits. Fable 5 remains the top of the range—the Mythos-class model for the most demanding reasoning and long-horizon autonomous work, priced at $10/$50 per million tokens on the API.

New: the Effort setting. The model picker now includes an Effort control (default: High). It determines how much thinking and work the model invests in a task: lower effort means faster, terser answers for routine work; higher effort buys deeper reasoning and more thorough verification. On the API the same control spans low, medium, high, xhigh, and max—with xhigh recommended for the hardest coding and agentic workloads.

One thing for API users to note: Sonnet 5 uses a new tokenizer that produces roughly 30% more tokens for the same text than Sonnet 4.6. Per-token pricing is unchanged, but token-based budgets, max_tokens limits, and cost dashboards calibrated against Sonnet 4.6 should be re-baselined. Subscription users don't need to do anything.

What should you pick? The rule of thumb barely changes, the names do: Sonnet 5 for everyday work, Haiku 4.5 when speed and volume matter, Opus 4.8 for complex multi-step tasks, and Fable 5 for the genuinely hardest challenges—especially while it's still included through July 12.