Enterprise

Anthropic launches Claude Sonnet 5 as the new default, pushing near-Opus performance at half the price

The most agentic Sonnet yet lands across Free, Pro, and every enterprise tier as Anthropic marches toward an IPO and enterprises fight ballooning agent bills.

By Asher Glick · Enterprise software desk · 2026-07-02 · 3 min read

Photo: Unsplash / Steve Johnson — Abstract neural network visualization representing an AI agent at work

Anthropic shipped Claude Sonnet 5 on June 30, 2026, promoting it to the default model across Free, Pro, and every enterprise tier, and priced it at an introductory $2 per million input tokens and $10 per million output tokens. That’s roughly 60% below Opus 4.8’s $5 and $25, per VentureBeat’s read, and the discount narrows to about 40% once standard pricing of $3 and $15 kicks in after August 31.

The pricing is the story, and the story is agents. Enterprises running long-horizon workflows have watched their token bills scale faster than the workflows themselves, and Anthropic is betting that a Sonnet tier priced within earshot of the frontier is what unsticks that market. The company says Sonnet 5 can plan, use browsers and terminals, and run autonomously at a level that until recently demanded larger, costlier models.

The benchmarks slot Sonnet 5 between its predecessor and Opus. It posts 63.2% on SWE-bench Pro against 58.1% for Sonnet 4.6 and 69.2% for Opus 4.8, and 80.4% on Terminal-Bench 2.1 versus 67.0% and 82.7%. On GDPval-AA v2, the knowledge-work test, Sonnet 5 scores 1,618, nudging past Opus 4.8’s 1,615. Anthropic also reports a lower rate of undesirable behaviors than Sonnet 4.6, which matters more in agentic deployments than in chat.

One asterisk: a revised tokenizer means the same input can map to between 1.0 and 1.35 times more tokens depending on content, so the sticker discount overstates the real one on some workloads.

Early customers are the proof point Anthropic wants in front of investors. TechCrunch reported that Zapier ran Sonnet 5 against a task it had previously botched. “the company handed Sonnet 5 a two-part job — updating Salesforce account tiers and sending a launch announcement to enterprise contacts — and it finished end to end, where the task used to stall halfway,” said Daniel Shepard, senior engineer at Zapier. Rakuten pointed it at dozens of its hardest production code pull requests.

VentureBeat notes Anthropic “barrels toward an initial public offering that will test whether the private market’s staggering AI valuations can survive public scrutiny.” That framing sits behind every decision here. Cheaper Sonnet, safer agents, wider default distribution: this is what a company builds when the next audience isn’t just developers but public-market investors reading the same benchmark table.

Sources