Anthropic ships Claude Sonnet 5.5: the same list price, up to 30% lower cost per task

The second Claude 5.5 model arrives at Sonnet 5's price

On a product page dated September 28, 2026, Anthropic presented Claude Sonnet 5.5 as the second release of its Claude 5.5 generation. It leads with two numbers: output generated more than 30 percent faster, and a cost up to 30 percent lower for most work, making it what Anthropic calls its fastest Sonnet model to date. The positioning is stated plainly: "Sonnet 5.5 is a faster, lower-cost complement to Claude Opus 5.5." In Anthropic's split, Opus 5.5 takes complex work that calls for careful judgment, while Sonnet 5.5 does its best work on clearly scoped day-to-day tasks, bug fixes, and polished documents, slides and spreadsheets.

TechCrunch covered the launch the same day under Lucas Ropek's byline. He recalls that Sonnet 5 was announced about three months earlier and sums up the pitch in one line: "The big selling point with 5.5, meanwhile, is speed." He also relays Anthropic's claim that the model burns through tokens at a markedly slower rate. According to Anthropic, Claude Haiku 5.5, aimed at high-volume, cost-sensitive applications, is due to join the family in the coming weeks; TechCrunch notes the company gave no firm date.

Unit prices hold, the saving comes from token counts

Anthropic keeps Sonnet 5.5's input, output and cache read rates where Sonnet 5 had them. Its pricing table sets Sonnet 5.5 next to Opus 5.5, which is listed higher for input, output and cache writes. The saving does not come from the rate card but from the model usually getting the same job done with far fewer tokens. Anthropic's own tests put the per-task saving over Sonnet 5 at a maximum of 30 percent, which is a ceiling, not a guarantee for every job. Effort settings shape the bill too: the default is Medium in Claude Code and Anthropic's apps and High on the Claude Platform. Anthropic also says that on several benchmarks, Sonnet 5.5 running at Low or Medium effort tops the best score Sonnet 5 managed, at roughly one tenth of the per-task cost.

The benchmark table does not point in one direction. On Terminal-Bench 4.0 Sonnet 5.5 scores 70.6 percent against Sonnet 5's 10.3 percent, with Opus 5.5 listed at 66.4 percent at Xhigh effort. On CursorBench 4.0 Opus 5.5 leads at 57.8 percent, ahead of Sonnet 5.5 at 55.5 and Sonnet 5 at 34.1. On GDPval-AA v2.1 the scores are 1844 for Sonnet 5.5, 1846 for Opus 5.5 and 1449 for Sonnet 5; those were run not by Anthropic but by Artificial Analysis, on a pre-release deployment carrying a since-fixed bug that could affect structured outputs. Anthropic expects any effect to be small and, if anything, to understate Sonnet 5.5. Anthropic itself says Opus 5.5 is still clearly the stronger model for complex, open-ended work that demands sustained judgment. The early testers' numbers are about efficiency. Abhinay Kathuria, Director of AI at Zendesk, reports that "Tickets were processed 20% faster". Slack's Curtis Allen says the model beat Sonnet 5 on almost all offline Slackbot evals while using about 14 percent fewer output tokens, and Lovable's Fabian Hedin saw a third fewer tool calls in its coding evals.

In agent work, tokens per task set the bill, not the rate card

In agent-based workflows, what drives cost is less the list price than the number of tokens a task burns before it is done. When the unit price holds and the token count falls, the budget for high-volume work such as a customer support bot or content automation can shift directly. Speed is a line item of its own: the faster a ticket closes, the less a customer waits. Still, the figures here are reported by Anthropic and its early testers, with the GDPval-AA scores run by Artificial Analysis, and the only way to know whether your workload sees the same result is to measure it.

The second dimension is architecture. Anthropic says the pairing with Opus 5.5 works best when Sonnet 5.5 runs at a lower effort setting, where each task is cheaper, which makes a split more reasonable: hard judgment calls go to the stronger model, routine steps to the faster one. The model is available across all platforms, Microsoft Azure, Google Cloud and Amazon Web Services among them, with the model ID claude-sonnet-5-5 on the Claude Platform, and like Opus 5.5 and Sonnet 5 it is offered with zero data retention. Teams that use Sonnet without thinking must first move to the new between_tools option before upgrading. Anthropic rates its cyber skills on par with Opus 5, so this is the first Sonnet release shipped with dedicated cyber safeguards: they resemble the ones on Opus 5.5, and riskier cybersecurity requests are visibly routed to Sonnet 5 instead.

The sources say nothing about Türkiye, so this section is commentary

Neither source mentions Türkiye. There is no measurement of Turkish-language performance, no price in Turkish lira, no Türkiye-specific availability condition and no customer example from the country. The reporting rests on Anthropic's product announcement and one technology outlet's coverage of it. What follows is therefore not a fact drawn from the sources but explicitly UNALSOFT commentary.

Our reading: because prices are quoted in dollars and the input, output and cache read rates did not change, the line that will move for a business using Claude through the API is the number of tokens each task consumes, and the only way to see that is to measure it on your own Turkish-language workload. The unknowns deserve the same clarity. The sources do not cover performance on Turkish text, pricing in Turkish lira, whether any Türkiye-specific availability condition applies, the exact date and price of Claude Haiku 5.5, or whether usage limits on free or paid consumer plans change. We covered the more capable model in the same family, Opus 5.5, separately; read together, the two pieces put both tiers of the family side by side.

Sources

Anthropic, Introducing Claude Sonnet 5.5 · TechCrunch, Anthropic releases Sonnet 5.5