Claude Sonnet 5 Becomes Anthropic Default Model at Lower Cost

Anthropic shipped Claude Sonnet 5 on June 30, 2026, and made it the default model for Free and Pro plans the very next day, with Max, Team, Enterprise, Claude Code, and the API platform all switched over at the same time. Pricing lands at $2 per million input tokens and $10 per million output tokens, an introductory rate that holds through August 31 before rising to $3 and $15. Opus 4.8, the flagship, costs $5 and $25. Sonnet 5 comes in at roughly 40% of that.

My read on this launch is that the pricing math matters more than the technical story. Sonnet 5 scores 63.2% on agentic coding against Opus 4.8’s 69.2%, a six-point gap for a 60% discount that most enterprise buyers can do in their heads. If you’re sitting on independent benchmark numbers, not Anthropic’s own, that don’t line up with 63.2%, I’d genuinely like to see them.

What Anthropic Is Betting On, One Day From Launch to Default

Anthropic’s official announcement lists improved agentic reasoning, tool use, and coding over Sonnet 4.6, plus the ability to complete tasks without explicit prompting, and lower hallucination and sycophancy rates. Zapier’s Daniel Shepard reported the model finishing multi-step tasks that used to require human intervention midway through. Lovable’s Fabian Hedin flagged consistent refusal of unsafe requests as the feature that matters most for a developer tool.

TechCrunch’s coverage places the launch in a wider pattern: Sonnet 5 undercuts Opus 4.8, GPT-5.5, and Gemini 3.1 Pro, and only Gemini 3.5 Flash comes in cheaper. The timing tracks OpenAI’s GPT-5.6 Sol and Google’s Gemini 3.5 Flash releases within the same stretch of weeks, with all three labs shifting the argument from raw capability to mid-tier cost efficiency.

Safety got an update too. Cyber safeguards are on by default, and Anthropic positions Sonnet 5’s capacity to develop cyberattack tools as substantially lower than Opus 4.8’s, alongside better resistance to prompt injection and malicious requests. The company is also upfront that Sonnet 5 doesn’t match Opus 4.8 on overall alignment metrics yet.

What the Numbers Actually Say

Start with the 63.2% figure itself. It comes from Anthropic’s own agentic coding benchmark, set next to Opus 4.8’s 69.2% and Sonnet 4.6’s 58.1% on the same table, produced by the same company running the same test suite. There’s nothing wrong with the number, but without independent confirmation from something like LMSYS Arena or Artificial Analysis, the table reads more like a well-built sales sheet than a neutral performance report.

The more interesting question is where this jump came from. Sonnet 4.6 to Opus 4.8 was an 11-point gap. Sonnet 5 closes that to 6 points against Opus while landing at 40% of the price. That kind of curve, performance closing in while cost drops by more than half, looks more like distillation or training-efficiency work squeezed into a smaller model than a genuine architectural leap. If this were architecture-level innovation, it would be a strange coincidence for the price to land precisely at 40% of Opus. That number reads like it came out of a pricing meeting, not a research lab.

Run the actual cost math and it gets clearer. A mid-sized agent task processing a million tokens costs roughly $12 on Sonnet 5 versus $30 on Opus 4.8, a 60% saving for giving up about 9% of relative performance (63.2 against 69.2). Scale that to an enterprise running a billion output tokens a month: Sonnet 5 runs about $10,000, Opus 4.8 about $25,000, a $15,000 monthly gap. That math, more than any benchmark headline, is why Sonnet 5 got promoted to default in a single day. A CFO looks at that spreadsheet and doesn’t need convincing about six percentage points of agentic coding.

VentureBeat’s reporting points to another layer: Anthropic is heading toward a high-profile IPO, and serving more users at lower marginal cost while improving gross margin is a natural move to make before going public. Pricing strategy tied to capital-markets narrative matters as much as any benchmark score when reading why this launched now.

Metrics Worth Watching

The first is independent verification. If LMSYS Arena or Artificial Analysis publish rankings within the next three to six months that land close to the claimed 63.2%, the performance story holds up. A meaningful gap would suggest the official number was built around a favorable comparison.

The second is whether the post-August 31 price increase actually happens. Going from $2/$10 to $3/$15 is a 50% jump, and whether Anthropic follows through if OpenAI or Google undercut it with a cheaper mid-tier model by then is worth watching.

The third is real enterprise migration speed. Early positive signals from Zapier and Lovable need months of production use to confirm, and the ratio of Sonnet 5 to Opus 4.8 API calls on Claude Code is a concrete number to track.

The fourth is how this launch lines up with Anthropic’s IPO timeline. If the S-1 process moves forward in the coming months alongside this cost-first pricing push, that combination will say more about the motive than the launch announcement does on its own.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

---

Related reading:

• Claude Opus 4.8: 41-Day Upgrade Cycle, Dynamic Workflows Run Hundreds of Parallel Agents
• Anthropic Files Confidential S-1: The $965 Billion AI IPO That Could Reset the Market
• OpenAI Eyes Major Price Cuts as Claude Code Forces an AI Token War