Anthropic released Claude Sonnet 5 today, positioning the model as a substantial upgrade for developers who need autonomous AI agents but cannot afford the company's flagship Opus pricing. The timing carries significant weight: Anthropic's most capable models remain offline following a US government export control order, and the company has filed confidentially for what could be one of the largest tech IPOs in years.
The Model
Sonnet 5 is built for agentic work. It can plan, use tools like browsers and terminals, and execute multi-step workflows autonomously. According to Anthropic, testers found the model finishes complex tasks where previous Sonnet versions would stall, and checks its own output without being prompted to do so.
On benchmarks, Sonnet 5 approaches Claude Opus 4.8 while costing substantially less. The company says it scores 63.2% on an agentic coding evaluation, compared to Opus 4.8's 69.2% and the previous Sonnet 4.6's 58.1%. For knowledge work tasks, Sonnet 5 actually edges out Opus 4.8 slightly.
Pricing starts at $2 per million input tokens and $10 per million output tokens through August 31, after which it rises to $3 and $15 respectively. Developers can access it through claude-sonnet-5 in the API. Free and Pro users on Claude.ai get it as their default model starting today.
Sualeh Asif, co-founder of Cursor, said the model "stay[s] on plan, follow[s] our conventions, and ship[s] clean multi-step changes, all at an efficient cost."
What Sonnet 5 Cannot Do
Anthropic's system card for Sonnet 5 is candid about its limits. The model does not advance the company's capability frontier compared to Opus or Mythos-class models. On cybersecurity evaluations, Sonnet 5 scored 0.0% on a Firefox exploit development test, far behind Opus 4.8's 68.8% and Mythos 5's 88.4%. Anthropic launched the model with cyber safeguards enabled by default.
The system card also notes that Sonnet 5 shows "somewhat higher rates of misaligned behavior" than Opus 4.8. Safety assessments found lower rates of undesirable behaviors than Sonnet 4.6, but the model is not on the same level as Opus when it comes to alignment.
The Fable 5 Shadow
Sonnet 5 arrives while Anthropic's most advanced models remain suspended. On June 12, three days after launching Fable 5 and Mythos 5 to considerable fanfare, the company received an export control directive from the US Commerce Department. Commerce Secretary Howard Lutnick ordered Anthropic to suspend access for any foreign national, anywhere in the world. Anthropic disabled both models globally because it could not filter users by nationality in real time.
The trigger, according to reports, was a jailbreak technique that could bypass Fable 5's safety guardrails. Anthropic contests the government's characterization, calling the jailbreak narrow and non-universal, and arguing that similar capabilities exist in competing models like GPT-5.5 that face no such restrictions.
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company wrote. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."
The suspension deepens tensions between Anthropic and Washington. Earlier this year, the Department of Defense labeled Anthropic a supply-chain risk to national security, a designation typically reserved for foreign adversaries. Litigation over that decision is ongoing.
What This Means for Sonnet 5
In practice, the Fable suspension makes Sonnet 5 more important than it would otherwise be. Anthropic's mid-tier model is not subject to the same restrictions. For developers who integrated Fable 5 into production last week and saw their traffic break overnight, Sonnet 5 offers a path forward that does not depend on geopolitical stability.
The IPO calculus matters here too. Anthropic filed its prospectus confidentially in early June, disclosing a $47 billion revenue run rate and a $965 billion valuation from its most recent funding round. A model good enough to rival Opus and cheap enough to run at scale could prove essential to the company's public-market narrative, especially if investors question whether Anthropic can stay at the frontier if Washington keeps pulling its best models offline.
For now, Claude Sonnet 5 is available. How long that remains true for any given Anthropic model is a question the company can no longer answer on its own.
