Anthropic has released Claude Sonnet 5, replacing Sonnet 4.6 as the default medium-sized model for both Free and Pro Claude users. Sonnet sits between Anthropic’s smaller Haiku model and its larger Opus model. The new release arrives four months after Sonnet 4.6 shipped in February, continuing a rapid cadence as Anthropic competes with OpenAI and Google for coding and agent supremacy.
Anthropic is positioning Sonnet 5 as the most agentic Sonnet model it has shipped, capable of making plans, using tools like browsers and terminals, and running autonomously at a level that previously required larger and more expensive models. The company frames the release explicitly as closing the gap with its flagship tier:
Sonnet 5 narrows the gap: its performance is close to that of Opus 4.8, but at lower prices. It’s a substantial improvement over its predecessor, Sonnet 4.6, on important aspects of agentic performance like reasoning, tool use, coding, and knowledge work.
What’s new in Sonnet 5
Anthropic says the focus of this release is agentic reliability rather than a single headline benchmark. That means longer task chains without losing context, better self-correction when a tool call fails, and steadier behavior when a session runs for an extended period inside Claude Code or Cowork. The company describes Sonnet 5 as better at deciding when to ask a clarifying question versus when to proceed on its own, which has been a common complaint with earlier agentic models that either stalled too often or charged ahead with wrong assumptions.
Anthropic also says the model handles multi-step coding tasks, like refactoring across several files or coordinating changes between a frontend and backend, with fewer dropped steps than Sonnet 4.6. For developers building on top of Claude, that translates into fewer retries and less manual cleanup after an autonomous run.
Pricing for Claude Sonnet 5
Claude Sonnet 5 is available everywhere today at an introductory price of $2 per million input tokens and $10 per million output tokens, a rate that holds through August 31, 2026. After that it moves to standard pricing of $3 per million input tokens and $15 per million output tokens. Anthropic has also increased rate limits across Chat, Cowork, Claude Code, and the Claude Platform to accommodate higher token usage at higher effort levels, letting users pick whichever effort level fits their project.
The pricing structure mirrors the strategy Anthropic has used with prior Sonnet releases: undercut Opus pricing significantly while closing the capability gap, so teams that previously had to pay for the larger model just to get reliable agentic behavior can now move down a tier without sacrificing much performance.
Mythos 5 and Fable 5: Anthropic’s higher capability models
Alongside Sonnet 5, Anthropic disclosed two more specialized variants. Mythos 5 carries fewer guardrails and is reserved for select security researchers and platform owners. Fable 5 is the safer counterpart, and it was briefly available to customers before the U.S. government blocked access. Anthropic says it is working on restoring that access in the future.
These two variants are not part of the general Sonnet 5 rollout and are not available through Chat, Cowork, or the standard Claude Platform pricing tiers. They exist primarily as research and security tooling rather than as products aimed at everyday developers or consumers.
Sonnet 5 is rolling out as the default model for Free and Pro users in Claude’s consumer apps, and it is also accessible through Claude Code and the Claude Platform for developers building their own tools and agents on top of Anthropic’s API. As with previous Sonnet releases, Anthropic typically makes the model available through major cloud partners as well, giving enterprise customers a path to adopt the new model without changing their existing infrastructure provider.
1 comment