Anthropic released Claude Sonnet 5 on Tuesday at $2 per million input tokens and $10 per million output tokens through August 31, roughly 60% under the standard Opus 4.8 rate of $5 and $25. The pricing sits inside a much larger choreography: a confidential IPO filing with the SEC in early June, a half-price statewide deal signed by California Governor Gavin Newsom the day before launch, and the ongoing rehabilitation of two Anthropic models the Commerce Department forced offline eighteen days earlier.
The company calls Sonnet 5 “the most agentic Sonnet model yet” and says it “can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models.” The benchmark numbers, via TechCrunch, back the pitch narrowly: 63.2% on an agentic coding evaluation, against 58.1% for Sonnet 4.6 and 69.2% for Opus 4.8. VentureBeat’s read is that Sonnet 5 “vaults into a performance tier that overlaps substantially with Anthropic’s flagship.” At two-fifths the price, it doesn’t need to close the gap. It only needs to make Opus a harder purchase order to justify.
Sonnet 5 becomes the default for Free and Pro users worldwide and slots into Claude Code as the standard agent. After August 31, pricing resets to $3 and $15. An updated tokenizer maps the same input to 1.0–1.35× more tokens depending on content type, which Anthropic describes as “roughly cost-neutral” once the introductory window closes. That framing won’t survive contact with anyone’s finance team.
The safety story is doing its own work. In Mozilla-partnered testing against Firefox 147, Sonnet 5 registered a zero percent success rate at generating a working exploit; Mythos 5, by VentureBeat’s account, succeeded nearly 90% of the time on the same test. Anthropic omitted specialized cybersecurity datasets from Sonnet 5’s training, per Artificial Intelligence News. The context matters: on June 12, a Commerce Department export control directive forced Anthropic to suspend Fable 5 and Mythos 5 after Amazon researchers documented a jailbreak that got Fable 5 to identify software vulnerabilities and supply exploitation code. Both models, alongside Opus 4.7, are now restored.
Read together, the week is a single message to two audiences. To enterprise buyers and California procurement officers: cheaper, safer, defaulted-in. To the SEC, the Pentagon, and a federal government that has labeled Anthropic a “supply-chain risk”: we can pull a model, retrain it, ship it back. CNBC calls the coming IPO “the most scrutinized offering of the year.” Anthropic is pricing the S-1 into the API.
Sources
- https://www.anthropic.com/news/claude-sonnet-5
- https://techcrunch.com/2026/06/30/anthropic-launches-claude-sonnet-5-as-a-cheaper-way-to-run-agents/
- https://techcrunch.com/2026/06/29/anthropic-and-gov-newsom-forge-deal-allowing-california-government-to-use-claude-at-half-price/
- https://www.artificialintelligence-news.com/news/anthropic-deploys-claude-sonnet-5-fable-and-mythos-restored/
- https://venturebeat.com/technology/anthropic-launches-claude-sonnet-5-at-a-steep-discount-to-its-top-model-as-the-company-races-toward-a-blockbuster-ipo