Anthropic launches Claude Sonnet 5 at a steep discount to its Opus flagship

Anthropic released Claude Sonnet 5 on June 30, 2026, pricing the mid-tier model at $2 per million input tokens and $10 per million output tokens through August 31. That is less than half the $5 and $25 that its Opus 4.8 flagship charges, and the rates rise to $3 and $15 after the introductory window.

The pricing is the news. Anthropic is undercutting its own top model and, on the company’s framing, positioning Sonnet 5 as the cheaper way to run software agents at volume. Anthropic calls it its most agentic Sonnet model to date, a pitch aimed at developers running long chains of tool calls where token costs compound fast.

On Anthropic’s own agentic-coding measure, Sonnet 5 scores 63.2 percent, above Sonnet 4.6’s 58.1 percent but still trailing Opus 4.8’s 69.2 percent. The company also reports a 0 percent success rate on a Firefox vulnerability-exploit test and lower rates of misaligned behavior than Sonnet 4.6. Those figures are Anthropic’s own and have not been independently verified.

Sonnet 5 became the default model for the Free and Pro plans at launch on June 30 and is available across the Free, Pro, Max, Team and Enterprise tiers. The wider distribution matters because it puts the cheaper model in front of Anthropic’s largest user base by default, rather than reserving it for paying developers.

The launch lands as Anthropic is reported to be moving toward an initial public offering, and Sonnet 5’s rates sit below OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro – a signal that price, not just capability, is now a front on which the frontier labs compete. Whether cheaper tokens translate into the agent workloads Anthropic is courting will show up in usage, not benchmarks.