Updated on September 15, 2026
Claude Sonnet 5 is Anthropic's mid-tier AI model, built for coding, autonomous agents and professional work at scale. Released on June 30, 2026, it comes close to Opus-class results on agentic tasks, with API rates of $2 per million input tokens and $10 per million output. On Claude.ai, it runs as the default model of the free plan, on the web as well as iOS and Android.
Claude Sonnet 5 is the most autonomous Sonnet Anthropic has ever shipped. The model drafts a plan, opens a browser, fires off terminal commands and chains steps without supervision, a level of independence that still required far pricier models a few months before its release.
Its context window reaches one million tokens with no long-context surcharge, and a single reply can stretch to 128,000 tokens. Where many large language models still cap out around 200,000 tokens, you can feed it an entire codebase. Adaptive thinking kicks in by itself depending on difficulty, and effort levels up to xhigh let you dose the compute budget.
In practice, you point it at a bug, it writes a test reproducing the issue, fixes it, then checks the failure comes back without the patch (the kind of reflex nobody expected from a machine). Early testers reported it finishes jobs where previous Sonnet models gave up.
On Anthropic's evaluations, Sonnet 5 sits close to Opus 4.8 for reasoning, tool use, coding and knowledge work. Against GPT-5.6 Terra or Gemini 3.1 Pro, both listed around $2 input and $12 output at OpenAI and Google, its rate card holds up well, and the LLM model rankings make it easier to track this neck-and-neck race between releases.
The debate is real, though. On Reddit, developers questioned whether the highest effort levels still make financial sense when Opus 4.8 matches them on some agentic search charts. The usual answer boils down to one word, routing.
Claude Sonnet 5 can be tried without a credit card, since it powers the free tier of Claude.ai by default. Pro, Max, Team and Enterprise subscribers get it too, as do Claude Code, Amazon Bedrock, Google Vertex AI and Microsoft Foundry on the enterprise side.
Through the API, the claude-sonnet-5 model ID costs $2 per million input tokens and $10 per million output, a rate announced as introductory at launch and still listed since. Prompt caching trims up to 90% off the bill and batch processing cuts it in half. These figures move with Anthropic's announcements, so treat the official pricing page as the final word.
| Anthropic model | API rate (input/output, per million) | Place in the lineup |
|---|---|---|
| Haiku 4.5 | $1 / $5 | High volume, lowest latency |
| Sonnet 5 | $2 / $10 | Daily coding and agents |
| Opus 5 | $5 / $25 | Heaviest reasoning |
Yes, partly. Sonnet 5 is the default model on the free Claude.ai plan, on the web and mobile apps, within limited usage quotas. The API is billed separately at $2 per million input tokens and $10 per million output tokens, regardless of any Claude Pro or Max subscription you may hold.
Sonnet 5 covers everyday execution, coding, agents and document analysis, at a far lower cost. Opus keeps the edge on the longest reasoning chains and high-stakes decisions. At maximum effort levels the price gap tightens, which is why many teams route each task to the right tier instead of picking one model forever.
It can. Anthropic reports that the same text produces roughly 30% more tokens than with Sonnet 4.6, depending on the workload. Per-token rates stay the same, but old cost estimates no longer apply. Recount your typical prompts before migrating, especially for agents that reread the same context over and over.
Enterprise deployments are a supported path. The model is available through Amazon Bedrock, Google Vertex AI and Microsoft Foundry, with regional data residency and enterprise-grade security controls. Anthropic's safety assessments also found fewer undesirable behaviors than Sonnet 4.6 in agentic settings.
Verdict: All the machinery of a high-end agent minus the high-end invoice, and that is what developers, analysts and teams automating their daily work at scale will remember.
GPT-6 Astra 
Claude Fable 5.1 
Claude Fable 5 
