Anthropic's Claude Sonnet 5, the mid-tier model the company built to make AI agents cheap enough for everyday production use, got 50% more expensive overnight. As of September 1, 2026, the model's introductory pricing lapsed and standard rates of $3 per million input tokens and $15 per million output tokens took effect, up from the $2/$10 rate Anthropic had held since Sonnet 5 launched on June 30, 2026.
The increase was baked into the launch terms from day one. When Anthropic introduced Sonnet 5, the company said the $2/$10 rate would hold "through August 31," after which pricing would jump to $3/$15. TechCrunch, which covered the launch, later appended a correction to its story clarifying the post-deadline output price was $15 per million tokens rather than an earlier figure — underscoring how closely developers were tracking the exact numbers months in advance. FinOps LLM, a cost-auditing firm that publishes research on enterprise AI spend, called the change a "hard deadline" in a report published in mid-July, modeling the impact for teams at scale: a workload running 500 million input tokens and 150 million output tokens a month would see its bill rise from $2,500 to $3,750, a $15,000 annual increase, once the calendar flipped to September.
Discount Ends, Standard Pricing Resumes
Sonnet 5 was pitched as Anthropic's answer to a market where every major lab was racing to make agentic capability the default, not the premium tier. "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," Anthropic said in its launch announcement, positioning Sonnet 5 as a cheaper alternative to its own Opus 4.8 as well as OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro. Early enterprise adopters credited the model with closing gaps that had previously required human intervention. "We handed Claude Sonnet 5 a two-part job — update Salesforce account tiers, send a launch announcement to enterprise contacts — and it finished end to end," said Daniel Shepard, a senior engineer at Zapier, in a statement accompanying the launch. "That used to stall halfway. For day-to-day automation, it's a no-brainer."
Fabian Hedin, co-founder of the app-building platform Lovable, praised the model's safety behavior in the same launch materials, noting that Sonnet 5 "refuses unsafe requests cleanly and consistently" — a distinction Anthropic has leaned on as it pushes agentic models into higher-stakes workflows like account management and outbound communications.
The pricing reset lands alongside a broader model-lifecycle reshuffling that has been running in the background all year. Anthropic's published model-deprecation policy commits to at least 60 days' notice before retiring any publicly released model, and the company has been working through that schedule methodically: Claude Sonnet 4 and Claude Opus 4 were deprecated on April 14, 2026 and retired exactly 60 days later, on June 15; Claude Opus 4.1 was deprecated June 5 and retired August 5. Anthropic's official documentation states plainly that "impacted customers will always be notified by email and in the documentation," a policy the company has now exercised on three separate model generations within a five-month span. Sonnet 5 itself carries a "not sooner than June 30, 2027" retirement floor on Anthropic's current model-status table — a full year of runway, but a reminder that even the newest model has a clock running.
Why It Matters
The Sonnet 5 price reset is a case study in a pattern now familiar across the foundation-model industry: launch a capable model at a loss-leading rate to win developer mindshare and default-model placement, then normalize pricing once the workloads are locked in. OpenAI and Google have followed similar playbooks with their own agentic-model launches this year — GPT-5.6 Sol and Gemini 3.5 Flash both arrived with aggressive introductory positioning before settling into standard tiers. For Anthropic, the timing is notable: the price increase landed the same week reports surfaced that investors are pushing for a roughly $2 trillion valuation ahead of a possible October IPO, built on projections that annualized revenue climbed from about $9 billion at the end of 2025 to more than $65 billion by the end of July 2026. A 50% price increase on one of the company's highest-volume models is a direct lever on that revenue trajectory, even as it raises costs for the developers and startups whose usage underpins those numbers.
The increase also arrives as Anthropic faces mounting legal exposure — Sony Music Publishing and Warner Chappell Music filed suit this week accusing the company of using pirated songs to train Claude, a complaint that names CEO Dario Amodei personally and covers a far larger catalog than prior music-industry claims against the company. None of that litigation risk shows up in the API pricing page, but it sits in the same news cycle as a company simultaneously asking developers to pay more and investors to pay dramatically more.
What to Watch
Expect FinOps teams at API-heavy shops to spend September auditing which workloads still justify Sonnet 5 at $3/$15 versus cheaper alternatives like Claude Haiku 4.5, GPT-5.6 Luna, or Gemini 2.5 Flash — several of which now undercut Sonnet 5 on cost per token by wide margins. Watch also for whether Anthropic offers volume commitments or negotiated rates to offset the increase for its largest customers, and whether rival labs use the moment to undercut Sonnet 5 on price to peel off cost-sensitive developers ahead of Anthropic's expected IPO push this fall.
“We handed Claude Sonnet 5 a two-part job — update Salesforce account tiers, send a launch announcement to enterprise contacts — and it finished end to end. That used to stall halfway.”— Daniel Shepard, Senior Engineer, Zapier