Anthropic Ships Claude Sonnet 5.5: 30% Faster, 30% Cheaper, Six Days After Opus 5.5

SAN FRANCISCO — Anthropic is not pacing itself. Six days after launching its flagship Claude Opus 5.5, the company has shipped the second model in the Claude 5.5 family: Claude Sonnet 5.5, a faster, leaner model aimed squarely at the everyday work that makes up the bulk of what developers and knowledge workers actually do. "It's a clear upgrade over Claude Sonnet 5, runs 30%+ faster, and costs up to 30% less for most work," the company said in its announcement.

If the cadence feels aggressive, that's because it is. Anthropic announced Opus 5.5 on September 22, promised Sonnet 5.5 and Haiku 5.5 would follow "in the coming weeks," and then dropped Sonnet less than a week later. In an industry where model releases used to be quarterly events with choreographed keynotes, six days between flagship and mid-tier reads like a company that has decided the release calendar is a weapon.

So what is Sonnet 5.5, exactly? Anthropic draws the line in one sentence: Opus 5.5 is built for complex work that needs careful judgment; Sonnet 5.5 is the lower-cost complement, strongest at well-scoped everyday tasks — fixing bugs, writing documents, building slides and spreadsheets. It's the workhorse model, the one that will quietly handle the other 90 percent of the calls hitting Anthropic's API on any given day.

Here's the part worth reading twice. The per-token price did not move: Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens — identical to Sonnet 5, with cache reads at $0.20 and the standard 50 percent batch discount. The "up to 30% less" figure is not a price cut. It's an efficiency claim: Anthropic says the new model uses fewer tokens and batches more tool calls to reach the same result, so the bill for the same task drops by up to 30 percent. Output also generates more than 30 percent faster than Sonnet 5. Same sticker price, smaller receipt — if Anthropic's numbers hold in the wild.

And the benchmarks are where this gets interesting. On Terminal-Bench 4.0, an agentic coding evaluation, Sonnet 5.5 scored 70.6 percent — against 10.3 percent for Sonnet 5, and, notably, 66.4 percent for Opus 5.5. Read that again: the cheaper model beat the flagship on Anthropic's own launch table, at half the per-token price. Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, so Sonnet 5.5 undercuts it by exactly half on both. On CursorBench 4.0, Sonnet 5.5 hit 55.5 percent versus 34.1 for Sonnet 5 and 57.8 for Opus 5.5 — closer to a dead heat. The honest read: for agentic coding, the mid-tier model is now the default, and Opus keeps a narrower job — the hard, ambiguous cases where a missed bug costs more than the tokens.

The specs round out the picture. Sonnet 5.5 carries a 1-million-token context window with up to 128,000 output tokens and a June 2026 knowledge cutoff, called via the API as claude-sonnet-5-5. It launched everywhere at once: the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure, plus the Claude apps — and GitHub began rolling it into Copilot the same day, which matters more than it sounds. When the model you write code with updates inside the editor you already live in, adoption isn't a migration; it's a Tuesday.

Early production feedback has been strong. Abhinay Kathuria, director of AI at Zendesk, said the model "made fewer wrong decisions and resolved tickets faster than the Claude models we use in production today." Anthropic also says Sonnet 5.5 is the first Sonnet to ship with cybersecurity safeguards similar to those on its most capable models — a quiet but telling detail, given how much of this model's workload will be agents touching real systems.

Zoom out and the timing looks less like a product launch and more like a chess move. OpenAI answered a day later at its DevDay with GPT-6.1 Sol — its own mid-tier model, at the exact same $2/$10 list price. Frontier mid-tier pricing is converging on $2/$10 per million tokens the way phone plans converged on unlimited: the number stops being a differentiator and the fight moves to speed, context, and whose benchmarks you believe. Anthropic's bet is that "30 percent faster and 30 percent cheaper per task" is the sentence that wins enterprise deals, not another leaderboard screenshot.

There is a third act still to come. Haiku 5.5 — the smallest, cheapest model in the family — is due in the coming weeks, and Anthropic has said it will carry many of the same gains in performance, efficiency, and safety. If Sonnet 5.5 is the workhorse and Opus 5.5 is the specialist, Haiku is the one that goes into everything: the background agents, the always-on copilots, the features where the model choice is measured in fractions of a cent.

The raised-eyebrow version of this story is simpler. A year ago, a model that beats its own flagship on agentic coding at half the price would have been a scandal or a mistake. In 2026, it's a Tuesday in September — and the real question isn't whether Sonnet 5.5 is good, it's what Anthropic ships next week. The company that releases flagship and mid-tier six days apart is telling you exactly how it plans to fight this war: not with one big model, but with a conveyor belt.

Developers don't need the philosophy. They need the routing rule, and Anthropic handed it to them: well-scoped work goes to Sonnet 5.5, judgment calls go to Opus 5.5. The AI infrastructure buildout keeps setting records, the labs keep shelving models over failed safety tests and losing the people in charge of them, and somewhere in the middle, the most important model of the fall might be the one that just does the boring work faster and cheaper. That's Sonnet 5.5. Boring, in this case, is the compliment.