Anthropic's Claude Sonnet 5 is cheaper, smarter, and finally ready for agents that run all day

Anthropic released Claude Sonnet 5 on June 30, 2026, a hybrid reasoning model for coding, agentic tasks, and enterprise deployments. The company calls it the fastest, smartest Sonnet yet, combining speed and intelligence with pricing that makes it viable for both real-time agents and high-volume production work. For context on how quickly the pricing landscape is shifting, see how Kimi K3 rewired AI economics.

The model's 1-million-token context window lets it work through long documents, follow complex multi-step instructions, and stay coherent over extended tasks. Anthropic emphasizes Sonnet 5's performance in advanced coding, long-running autonomous agents, and computer use, positioning it as a backbone for everything from customer-facing bots to internal automation. The agentic push aligns with findings that working memory limits agent reliability at long tasks.

Pricing and availability

Sonnet 5 is live on Claude.ai (web, iOS, Android), the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. Through August 31, 2026, introductory pricing sits at $2 per million input tokens and $10 per million output tokens. After that, standard rates kick in: $3 per million input tokens and $15 per million output tokens. Customers can cut costs by up to 90% with prompt caching and 50% with batch processing, a draw for high-volume use cases. Organizations that need US-only data processing can opt for a 1.1x pricing tier on both input and output tokens.

Benchmark performance

Anthropic says Sonnet 5 posts strong numbers across reasoning, tool use, software coding, and professional knowledge work, though it did not release specific scores in the announcement. The company says the model went through extensive testing and evaluation, including collaboration with outside experts, and has published a system card detailing safety results.

Customer adoption and use cases

The announcement includes nods from several customers. Cognition Labs, which builds AI agents for software engineering, reports that Sonnet 5 handles sustained coding, tool use, and debugging well, even in messy technical environments. Growrk says a two-part job, updating Salesforce account tiers and sending a launch announcement to enterprise contacts, ran end-to-end without hiccups. Lovable highlights the model's ability to refuse unsafe requests cleanly, calling it a key trust and safety feature. Other customers praise the model for carrying real pull requests through to tested, verified results on its own, and for its performance on brownfield code, race conditions, hidden tests, and legacy systems. Legal tech firm Eve found clear gains in legal research and analysis. ClickHouse notes faster time-to-insight for live data queries. Pace, which uses computer-use agents for insurance workflows, emphasizes the model's speed and consistent decision-making.

The distinct safety approach, clean refusal without broken workflows, is rare enough to stand out. For a deeper look at why consistent refusal matters in deployment, see the uncensored model paradox.

Context: The Sonnet lineage

Sonnet 5 arrives after a rapid release cadence: Sonnet 4 in May 2025, Sonnet 4.5 in September 2025, and Sonnet 4.6 in February 2026. Each version brought improvements in coding, agentic capabilities, and domain knowledge. Sonnet 4.5 was hailed as the best model for agents, coding, and computer use at the time. Sonnet 5 extends that trajectory with a bigger context window, stronger reasoning, and lower cost. The speed of iteration mirrors what Groq's agent operating system aims to support across different model providers.

With its blend of performance, pricing, and safety testing, Claude Sonnet 5 is Anthropic's most ambitious push to bring agentic AI into production use at scale.