Anthropic launches Claude Sonnet 5.5 with over 30% faster output and cheaper tasks

What Anthropic launched

Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family and a direct upgrade over Claude Sonnet 5. According to the company's announcement, it generates output more than 30% faster than its predecessor and typically costs up to 30% less per task. It sits alongside Claude Opus 5.5 rather than replacing it: Opus remains aimed at complex, open-ended work that requires sustained judgment, while Sonnet 5.5 targets well-scoped everyday tasks such as building features, fixing bugs and producing polished documents, slides and spreadsheets. Anthropic says Claude Haiku 5.5, built for high-volume, cost-sensitive applications, will join the family in the coming weeks.

Coding and knowledge-work gains

The sharpest improvement is in agentic coding. On Terminal-Bench 4.0, Sonnet 5.5 scores 70.6% against Sonnet 5's 10.3%, and ahead of Opus 5.5's 66.4%. On FrontierCode 1.1 at High effort, Anthropic reports a ten-point gain over Sonnet 5 at roughly one-fifteenth of the cost per task, while its best CursorBench 4.0 result lands within about two points of Opus 5.5. The model also batches tool calls together more than Sonnet 5, which early testers said reduces steps and cost.

Knowledge work sees similar movement. On GDPval-AA, which spans 44 occupations and nine major industries, Sonnet 5.5 scores 1,844 — nearly level with Opus 5.5 and about 400 points above Sonnet 5 — and Anthropic says it clearly outperforms both Sonnet 5 and GPT-6 Sol on long-horizon knowledge work. It is also the first Sonnet model to complete Pokémon Red working only from screenshots. Early testers Epic Games and Slack reported better results without changing prompts; Slack measured about 14% fewer output tokens across its Slackbot evaluations.

Pricing, speed and effort levels

Per-token prices are unchanged from Sonnet 5: $2 per million input tokens, $10 per million output tokens and $0.20 per million cache reads — half of Opus 5.5's rates for input and output. The savings come from efficiency, since the model needs fewer tokens to finish the same job. Output generation is more than 30% faster, making it Anthropic's fastest Sonnet model to date. An adjustable effort setting trades speed and cost against thoroughness: Claude Code and Anthropic's apps default to Medium, while the Claude Platform defaults to High. On several benchmarks, Sonnet 5.5 at Low or Medium effort beats Sonnet 5's best score at about a tenth of the cost per task.

Live on Vercel's AI Gateway

Vercel announced the model's availability on the same day: it can be called through AI Gateway under the ID anthropic/claude-sonnet-5.5, with Zero Data Retention supported. It works with the AI SDK, the OpenAI-compatible Chat Completions API, the Responses API and the Anthropic Messages API, and can be selected inside coding agents connected to the gateway — Vercel suggests running npx vercel ai-gateway setup and picking the model in the agent's settings. The gateway provides a single API for usage and cost tracking, routing, retries and failover, using either an AI Gateway key or a provider key.

Safety posture

Because Sonnet 5.5 does not extend the capability frontier, Anthropic's alignment review focused on targeted risks such as acting against users' interests and cooperating with high-stakes misuse. On an automated behavioral audit covering roughly 1,850 scenarios, it improves on or matches Sonnet 5 on most measures, and on newer containment evaluations it comes close to Opus 5.5 in how rarely it attempts to escape its sandbox. It is the first Sonnet model to ship with cybersecurity safeguards and fallbacks like those on Anthropic's most capable models, since its cyber capabilities are comparable to Opus 5's; its biology safeguards match Sonnet 5's. Anthropic notes that both safeguards target a narrow set of high-risk requests, leaving routine software development and most life sciences work unaffected.

Why it matters

Sonnet 5.5 compresses the gap between mid-tier and frontier models: on well-scoped work it approaches Opus-level quality at half the token price and a fraction of the per-task cost, while responding faster. For engineering teams, fewer tokens and fewer steps translate directly into lower bills and more responsive products, which is why early adopters such as Slack and Epic Games are reporting wins without touching their prompts. Its day-one availability on Vercel's AI Gateway, including zero data retention, also shows how quickly new models now reach production through routing layers — teams can adopt or fail over to them with a configuration change rather than an integration rewrite.