Claude Sonnet 5

Claude Sonnet 5 is the most agentic Sonnet-class model to date — released by Anthropic on June 30, 2026 as the direct successor to Sonnet 4.6 (February 2026). It can plan, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models.

Position in Anthropic's portfolio

The Sonnet line started the agentic-AI era for developers (Sonnet 3.5, 3.6, 3.7 were the first models with impressive coding and tool-use skills). Recent clear gains in agentic capabilities came mostly from the Opus class — Sonnet 5 closes that gap: performance close to Opus 4.8 at lower prices. Sonnet 5 is the default model for Free and Pro plans; also available for Max, Team and Enterprise as well as in Claude Code and on the Claude Platform.

Key improvements

Substantial gains over Sonnet 4.6 in: reasoning, tool use, coding, and knowledge work. On the BrowseComp (agentic search) and OSWorld-Verified (computer use) agentic evaluations, Sonnet 5 is a strict improvement over Sonnet 4.6, and the configurable 'effort' level lets you balance cost and quality per request. A new tokenizer (as in Opus 4.7) changes the text-to-token mapping at a 1.0-1.35× ratio versus Sonnet 4.6 — introductory pricing has been set so the transition to Sonnet 5 is cost-neutral.

Safety

Pre-deployment safety: overall lower rates of misaligned behavior than Sonnet 4.6, though higher than Opus 4.8 and Mythos Preview. Fewer hallucinations and less sycophancy. Cyber: on software-exploit development evaluations (e.g. Firefox 147) Sonnet 5 was never able to develop a working exploit; slightly higher partial-success rate than Sonnet 4.6 — a side effect of general intelligence improvements rather than targeted training. Cyber safeguards enabled by default (the same as in Opus 4.7/4.8).

Pricing and availability

Introductory pricing (through August 31, 2026): $2/M input tokens and $10/M output tokens. Standard pricing (from September 1, 2026): $3/M input and $15/M output. Prompt caching: up to 90% savings; batch processing: 50% savings. US-only inference: 1.1× standard price. Availability: claude.ai (web/iOS/Android), Claude Code, Claude Platform, AWS Bedrock, Google Cloud Vertex, Microsoft Foundry.