Anthropic Monday released Claude Sonnet 5.5, a mid-tier model it says produces output more than 30% faster than Sonnet 5 and completes tasks for up to 30% less, without changing per-token prices. The launch, also covered by TechCrunch and VentureBeat, comes about three months after Sonnet 5 shipped in June.

What Happened

Sonnet 5.5 is available through the Claude API under the identifier claude-sonnet-5-5 and on Amazon Web Services, Google Cloud and Microsoft Azure, with a zero data retention option. Pricing is $2 per million input tokens and $10 per million output tokens, with cache reads at $0.20 and cache writes at $2.50 per million. VentureBeat noted those rates are unchanged from Sonnet 5; Anthropic lists Opus 5.5 at double the input and output prices.

The cost-per-task claim rests on efficiency rather than a price cut. According to Anthropic and VentureBeat, the model needs fewer tokens and fewer tool calls to finish the same work, and it generates output faster.

RelatedAnthropic Prices Claude Haiku 5.5 at a Tenth of Its Predecessor for Most API Requests

Anthropic's published scores show large gains on agentic coding. Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5 and 66.4% for Opus 5.5. It posted 1844 on GDPval-AA v2.1, a measure of economically valuable knowledge work, against 1449 for Sonnet 5 and 1846 for Opus 5.5. Other results included 55.5% on CursorBench 4.0 and 46.2% on FrontierCode 1.1 at maximum effort, both below Opus 5.5. All figures are the company's own.

Why It Matters

Sonnet sits between Opus, Anthropic's most capable line, and the smaller Haiku, and it is the tier many developers use for everyday coding agents and document work. A model that approaches Opus on several tests at half the list price changes the math for teams running long agent sessions, where token counts and tool calls, not the per-token rate alone, drive the bill.

Pricing competition is tight. VentureBeat calculated that Sonnet 5.5's combined input and output price matches OpenAI's GPT-6 Sol, while Google's Gemini 3.8 Flash is cheaper through the end of 2026.

New Cyber Restrictions

The release also moves Sonnet into stricter safety territory. Anthropic said the model's cyber capabilities are comparable to Opus 5, and TechCrunch reported it is the first Sonnet model subject to the same cyber safeguards as the Fable and Opus lines. VentureBeat reported that certain high-risk cybersecurity requests fall back automatically to Sonnet 5, while Anthropic said routine debugging is unaffected.

Anthropic also said Sonnet 5.5 is the first Sonnet with classifiers meant to stop users from extracting its reasoning, a defense against distillation, where competitors train their own models on another model's outputs.

Epic Games, Every and CodeRabbit are among the customers quoted in the announcement. Epic Games said the model met a quality bar it would expect from a higher-tier model on system architecture reviews.

What to Watch

Anthropic said Claude Haiku 5.5, aimed at high-volume, cost-sensitive uses, will join the 5.5 family in the coming weeks, without giving a date. Independent benchmark runs and developers' real-world bills will show whether the claimed 30% savings per task hold up outside Anthropic's tests, and how often the cyber fallback to Sonnet 5 affects legitimate security work.