Claude Sonnet 5: Opus-class Agentic Power at Sonnet Prices
Claude Sonnet 5, released June 30, 2026, closes the gap with Opus 4.8 on multi-step autonomous tasks — finishing workflows, self-checking output, and recovering from errors without hand-holding. Learn when and how to put its follow-through to work.
Claude Sonnet 5 is now the default model on Free and Pro plans. Its defining upgrade over Sonnet 4.6: it finishes multi-step tasks where previous Sonnet models stopped short — and checks its own work without being told to.
Until Sonnet 5, true agentic capability required Opus-class models. Sonnet 5 narrows that gap. Its performance on reasoning, tool use, coding, and knowledge work is close to Opus 4.8, at a lower price: $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3/$15 at standard pricing. Access it via the API as claude-sonnet-5.
The clearest signal from early testers: Sonnet 5 doesn't need you to micromanage its process. One engineer described asking it to investigate a bug — unprompted, it wrote a reproducing test, implemented the fix, then stashed the fix to confirm the bug returned without it. All in a single pass.
Another team handed it a two-part job — update Salesforce account tiers, then send a launch announcement to enterprise contacts — and it finished end to end. That workflow used to stall halfway with Sonnet 4.6.
This changes how you should prompt. Give Sonnet 5 the full goal, the context it needs, and the expected output. Skip the step-by-step handholding between stages.
Two safety improvements matter for agentic work. First, Sonnet 5 is better at refusing malicious requests and resisting prompt injection — important if you deploy it in pipelines that process untrusted input. Second, it shows lower hallucination and sycophancy rates than Sonnet 4.6, which makes its self-verification more trustworthy.
One practical note: Sonnet 5 uses an updated tokenizer. The same input can map to 1.0–1.35× more tokens depending on content type. Anthropic set introductory pricing to keep the transition roughly cost-neutral.
What you can now do with Sonnet 5: hand it a complete multi-step goal and expect it to follow through — including self-checking — without step-by-step prompting. It performs close to Opus 4.8 on agentic tasks at a lower price ($2/$10 per MTok introductory through Aug 31, 2026). Access it as claude-sonnet-5. For best results: give the full goal, the relevant context, and the expected output format. Let it plan and execute.