Anthropic launches Claude Sonnet 5 with expanded agentic capabilities and platform support

Anthropic announced the release of Claude Sonnet 5, positioning it as "our most agentic Sonnet yet" and highlighting its ability to perform planning, browser and terminal tool use, and autonomous execution that previously required larger models. The model is now the default mid-tier option in Claude Code for Pro users and is available across the Claude Platform, including API and Managed Agents.

The company introduced a 1 million-token context window for Sonnet 5 and maintained standard list pricing at $3 per million input tokens and $15 per million output tokens, while offering a promotional rate of $2 per million input tokens and $10 per million output tokens through August 31 or September 1, depending on the announcement.

Sonnet 5’s agentic capabilities were a focal point, with third-party evaluations noting improvements over its predecessor but questioning its price-performance relative to higher-tier models. Cursor reported a 57% score on CursorBench for Sonnet 5, compared to 49% for Sonnet 4.6, while Cognition reported that Sonnet 5 outperformed Opus 4.8 on FrontierCode Extended with a 53.8% score and a 57.6% pass rate.

Independent analysis from Artificial Analysis indicated that Sonnet 5 used approximately 69,000 output tokens per task on average, about 40% more than Sonnet 4.6, and cost roughly $2.29 per Intelligence Index task at standard pricing—about twice the cost of Sonnet 4.6 and roughly 15% more than Opus 4.8. The model also demonstrated a 3x increase in agentic turns over Sonnet 4.6 on specific benchmarks.

Alongside the model release, Anthropic expanded platform support, including the addition of Linux desktop support (Ubuntu/Debian beta) for Claude Desktop on paid plans, though Computer Use was not included in this release. The company also shipped updates to Managed Agents, introducing features such as streaming session deltas, per-session overrides, webhook events, reverse pagination, credential injection scoping, and an observability tab with token and tool metrics.

The launch followed a period of speculation about the simultaneous release of Sonnet 5 and Fable 5, with some users expecting Fable 5 to be reintroduced after a period of restricted access. However, Fable 5’s re-release was not announced alongside Sonnet 5, and discussions noted that Sonnet 5 had been released without the anticipated Fable 5 update.