Sonnet Catches Up to the Flagship: Claude Sonnet 5 Is Here
Anthropic's new Sonnet is positioned less as a chat assistant and more as something that gets the job done on its own. Claude Sonnet 5 handles planning, tool use, and autonomous operation that used to need larger models — and it checks and corrects its own output without being asked. Anthropic calls it "the most agentic Sonnet model yet."
In brief
- What's new vs Sonnet 4.6: stronger reasoning, tool use, coding, and knowledge work; it finishes multi-step tasks instead of stopping short, and reviews its own output.
- Performance: close to Opus 4.8, at a much lower cost.
- Pricing: introductory through Aug 31, 2026 — $2 per 1M input tokens / $10 per 1M output; standard $3 / $15.
- Access: default on Free and Pro; available to Max, Team, Enterprise; in Claude Code and the Claude Platform; via the API as
claude-sonnet-5. - Safety: lower misalignment rates than Sonnet 4.6; cybersecurity capability deliberately cut relative to Opus models, with cyber safeguards on by default; better at refusing malicious requests and resisting prompt injection.
Our take
Two things stand out. On price: Sonnet 5 does work close to last generation's flagship for far less — a pattern that now repeats with every Anthropic release. On timing, though, the picture is different. Days after Fable 5 was pulled by an export control, Anthropic is framing its new model around "less cyber capability than Opus, safeguards on out of the box." It's easy to read that as pure marketing — but for anyone wiring agents into real workflows, "how smart is it" and "what limits does it run inside" now carry the same weight.
Kaynak: Anthropic