Anthropic released Claude Sonnet 5 on Tuesday at introductory API pricing of $2 per million input tokens and $10 per million output tokens through Aug. 31, making it the default model for Free and Pro users. Standard pricing settles at $3 and $15 after that window, still roughly 40 percent cheaper per token than the company’s Opus 4.8 flagship at $5 and $25, and about 60 percent cheaper during the introductory promo.

The pricing choreography matters because Anthropic filed confidentially with the U.S. Securities and Exchange Commission on June 1, followed roughly a week later by OpenAI. What the S-1 draft needs to show, in the current environment, isn’t a bigger model but a cheaper one that enterprise buyers will actually run in production loops.

Anthropic pitches Sonnet 5 as “the most agentic Sonnet model yet.” The benchmarks back the framing selectively: 63.2 percent on SWE-bench Pro against 58.1 for Sonnet 4.6 and 69.2 for Opus 4.8; 80.4 percent on Terminal-Bench 2.1, below Opus 4.8 at 82.7 but well above the prior Sonnet at 67.0. On the Firefox 147 exploit development evaluation, Sonnet 5 landed a 13.2 percent partial-success rate, while Mythos 5 hit 88.4 percent, a reminder of where the frontier still lives.

Customer testimonials arrive on cue. Sualeh Asif, co-founder of Cursor, said “with Claude Sonnet 5, agents stay on plan, follow our conventions, and ship clean multi-step changes, all at an efficient cost.” Daniel Shepard, a senior engineer at Zapier, described Salesforce workflows the model now finishes: “That used to stall halfway.”

The updated tokenizer expands input by 1.0 to 1.35 times depending on content type, quietly clawing back some of the discount. Fable 5 returns globally July 1. The message to public-market underwriters is legible: unit economics, not scaling laws.

Sources