Anthropic just released Claude Sonnet 5 and, quietly but decisively, priced it like it intends to win enterprise adoption before the IPO spotlight fully locks in. Sonnet 5 becomes the default model for users on Anthropic’s Free and Pro plans, and it’s also available to Max, Team, and Enterprise customers. The company is also calling it the “most agentic Sonnet model yet,” with agentic workflows that can plan, use tools like terminals or browsers, and execute multi-step tasks more reliably than earlier Sonnet generations.

The pricing is the other half of the playbook, and it is designed to sting the status quo. Introductory API pricing is set at $2 per million input tokens and $10 per million output tokens through August 31, after which it rises to $3 and $15. That keeps Sonnet 5 far below Anthropic’s top-of-the-line Opus 4.8 pricing of $5 per million input tokens and $25 per million output tokens, while still making Sonnet 5 compete on results. Anthropic’s framing is clear: deliver near-flagship performance at mid-tier prices, then pull cost-conscious developers into “agentic” production use cases right as the company barrels toward an IPO.

Benchmarks are where this move tries to earn its keep. On SWE-bench Pro, an agentic coding benchmark, Sonnet 5 scores 63.2% versus Sonnet 4.6’s 58.1%, narrowing the gap toward Opus 4.8 at 69.2%. On Terminal-Bench 2.1, another coding evaluation, Sonnet 5 posts 80.4% compared with Sonnet 4.6’s 67.0% and Opus 4.8’s 82.7%. In multidisciplinary reasoning on Humanity’s Last Exam, Sonnet 5 scores 43.2% without tools and 57.4% with tools, essentially matching Opus 4.8’s 57.9%. On computer use tasks evaluated through OSWorld-Verified, it reaches 81.2%, up from 78.5%. And on the knowledge-work benchmark GDPval-AA v2, it scores 1,618, surpassing Opus 4.8’s 1,615 and far exceeding Sonnet 4.6’s 1,395.

Put differently: Sonnet 5 doesn’t just edge forward. It overlaps substantially with Anthropic’s flagship tier across multiple categories. Anthropic says the overlap comes with a cost advantage, describing token pricing as roughly 60% less at standard pricing and even less during the introductory period. That matters for enterprise buyers because agentic systems are not just “better chat.” They are deployed into workflows that can consume compute and tools repeatedly while attempting to finish a task. If a model is strong enough to complete more of the workflow reliably, the economics of automation improve. The source also ties this directly to early enterprise partners who describe reliability improvements rather than just nicer outputs.

Agentic reliability is the “production gate” for a lot of companies, and the launch leans hard into that. Sualeh Asif, co-founder of Cursor, said that “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 a two-part automation job, updating Salesforce account tiers and sending a launch announcement, saying it “used to stall halfway” with previous models but now completes end to end. Those details point to why this release is more than a model refresh: enterprises have been cautious about moving agentic AI from pilot projects to production because partial completion can be worse than failure, leaving workflows stuck and teams to babysit systems.

There is also a technical footnote that executives should not ignore. Sonnet 5 uses an updated tokenizer similar to the change introduced with Opus 4.7, which changes how the model processes text. Anthropic says the tradeoff can map the same input to roughly 1.0 to 1.35 times as many tokens depending on content type. Anthropic claims introductory pricing is calibrated to make the transition “roughly cost-neutral,” but high-volume workloads will still want to benchmark specific use cases carefully before assuming bills remain stable. This is the kind of second-order issue that doesn’t show up on a leaderboard, but it can show up in procurement reviews and margin conversations.

Safety disclosures add nuance rather than a simple “better” headline. Anthropic reports that Sonnet 5 has lower rates of hallucination and sycophancy than Sonnet 4.6, is better at refusing malicious requests, and is more resistant to prompt injection attacks in agentic contexts. In an automated behavioral audit, Sonnet 5 scored lower (meaning safer) overall than Sonnet 4.6. However, it also showed “somewhat higher rates of misaligned behavior” compared with the more capable Opus 4.8 and the company’s Claude Mythos Preview, a cybersecurity-focused model with tighter restrictions. On a Firefox 147 exploit development evaluation created with Mozilla, neither Sonnet model could develop a working exploit, with both scoring 0.0%, though Sonnet 5 had a slightly higher partial success rate (13.2%) than Sonnet 4.6 (8.8%). Both remain far below Opus 4.8 at 68.8% working exploits and Mythos 5 at 88.4%. Because of incremental gains in cyber-adjacent capabilities, Sonnet 5 launches with cyber safeguards enabled by default. Those safeguards mirror Opus 4.7 and 4.8 but are less restrictive than those applied to Fable 5, which Bloomberg reported on June 10 is blocked from responding to queries related to cybersecurity and biology. Organizations enrolled in Anthropic’s Cyber Verification Program automatically receive the same access on Sonnet 5 without reapplying.

All of this lands at a very specific time in Anthropic’s corporate arc. The company confidentially filed its IPO prospectus with the SEC in early June, setting up what CNBC described as “the most scrutinized public offering in tech history.” Anthropic’s funding and revenue trajectory has been staggering: in February it raised $30 billion at a $380 billion valuation and reported $14 billion in annualized revenue that had “grown more than tenfold in each of the past three years,” as By late May it closed a $65 billion Series H round at a $965 billion post-money valuation, co-led by Altimeter Capital, Sequoia Capital, and others, with revenue run rate crossing $47 billion. PitchBook’s Harrison Rolfes told CNBC that the number that will “either validate or collapse the entire narrative the private markets have been pricing for three years” won’t be valuation or revenue, but gross margin, a figure no outside observer has yet seen.

In that context, Sonnet 5 is doing double duty. For developers, it offers capability improvements at competitive prices and expands the pool of users who can access agentic workflows. For Anthropic’s IPO story, it potentially supports the argument that distribution and compute economics can scale with a tiered product line, rather than forcing everything through the most expensive model. For peers, the stakes are simple: if Sonnet 5’s cost-performance overlap is real in production, enterprise buyers will treat “default model choice” and “token economics” as board-level issues, not just engineering preferences. And as the IPO approaches, that shift is exactly the kind of traction story public markets tend to reward, or punish, depending on gross margin follow-through.