Released 28 September 2026

Claude Sonnet 5.5Sonnet price, near-Opus coding

Claude Sonnet 5.5 is the second model in Anthropic's Claude 5.5 family, released on 28 September 2026. It scores 70.6% on Terminal-Bench 4.0 — ahead of Opus 5.5's 66.4% — while keeping Sonnet pricing at $2 per million input tokens and $10 per million output tokens. It runs on Felo AI Search and the Felo API.

Live since 28 September 2026 at $2 / $10 per million tokens, with output 30%+ faster than Sonnet 5.

Try Claude Sonnet 5.5 on Felo Call via Felo API

Released 28 September 2026 · Available now on Felo AI Search and the Felo API

At a glance

Claude Sonnet 5.5

Model ID
claude-sonnet-5-5
Released
28 September 2026
Input
$2 per million
Output
$10 per million
Cached input
$0.20 per million
Context window
1,000,000 tokens

Prices are Anthropic's published standard API rates per million tokens. Prompt caching cuts input cost by up to 90% and batch processing by 50%.

70.6%

Terminal-Bench 4.0

Agentic coding, ahead of Opus 5.5's 66.4%.

55.5%

CursorBench 4.0

Two points behind Opus 5.5, at a fraction of the cost.

1M

Context window

1M tokens in, up to 128K tokens out.

30%+

Faster output

Than Sonnet 5, the fastest Sonnet yet.

Benchmarks

Where Sonnet 5.5 lands

Anthropic's launch figures, with Sonnet 5 and Opus 5.5 alongside for scale. Sonnet 5.5 closes most of the gap to Opus 5.5 and passes it on agentic terminal coding.

Benchmark

Sonnet 5.5

Sonnet 5

Opus 5.5

GPT-6 Sol

Terminal-Bench 4.0

70.6%

10.3%

66.4%

—

FrontierCode 1.1

52.1%

42.4%

54.4%

49.3%

CursorBench 4.0

55.5%

34.1%

57.8%

—

GDPval-AA v2.1

1844

1449

1846

1487

OSWorld 2.1

80.1%

57.0%

81.8%

—

Chartography

61.6%

15.6%

64.4%

53.6%

Anthropic launch figures, September 2026. Terminal-Bench 4.0 and CursorBench 4.0 are reported at the model's best effort level; FrontierCode 1.1 is shown at xhigh, where Sonnet 5.5 scores 52.1%. GDPval-AA and AA-Briefcase are Elo scores, not percentages.

Source: Anthropic, Introducing Claude Sonnet 5.5, 28 September 2026

Cost

Sonnet pricing, fewer tokens per task

Sonnet 5.5 keeps Sonnet 5's list price but needs fewer tokens to finish the same work. Anthropic puts the saving at up to 30% per task for typical workloads.

Price per million tokens

Standard API rates, as published.

Input
$2
Output
$10
Cache read
$0.20
Cache write
$2.50

Prompt caching cuts input cost by up to 90%, and batch processing by 50%. US-only inference is billed at 1.1x.

Tokens per task, measured by customers

Three published results show the token drop on real workloads.

Balyasny Asset Management

On 2,441 finance tasks, Sonnet 5.5 used about 121,000 tokens per answer where Sonnet 5 used 497,000 — roughly 75% fewer.

Box

2.4x faster than Sonnet 5, with 12% fewer total tokens, and it caught errors in source documents that Sonnet 5 missed.

Slack

Better results on almost all offline Slackbot evals, in fewer steps and with about 14% fewer output tokens.

Figures are the customers' own published results on Anthropic's launch page, not Felo benchmarks.

Effort control

Five effort levels, two defaults

Effort is the main dial for how long Sonnet 5.5 thinks, and with it quality, latency and cost. The levels are recalibrated from Sonnet 5, so a fresh sweep against your own evals is worth running.

Low

Fastest and cheapest. Skips thinking on most simple requests; may skip verifying a code change.

Medium

The default in the Claude apps. A balance for chat and well-specified agentic tasks.

High

The default on the Claude API. The starting point for harder coding and multi-step tool use.

Xhigh

Extra thoroughness. Sonnet 5.5's best FrontierCode score, 52.1%, is at this level.

Max

The most thinking. Reserve it for work where you have measured a quality gain.

On FrontierCode, Max scored lower than Xhigh: the model ran a code-review skill that split work across subagents and, in two cases, timed out or edited beyond the task scope.

Capabilities

What Sonnet 5.5 is good at

Anthropic positions Sonnet 5.5 for well-scoped everyday work, with Opus 5.5 still the pick for complex, open-ended problems.

Agentic coding

Understands a codebase quickly, scopes changes before starting, and keeps edits small enough to review. Built for building features, fixing bugs and reviewing code.

Agents and tool use

Anthropic's strongest Sonnet yet on long-horizon tasks, and it batches tool calls to cut steps and cost. Lovable measured a third fewer tool calls and roughly half the shell runs.

Design and visual work

Adds polish to user interfaces, builds structured diagrams and user flows, and follows slide templates to produce decks that need minimal editing.

Knowledge and office work

Clearer, more direct writing across analysis, documents and slides. On GDPval-AA it scores 1844, within two points of Opus 5.5's 1846.

Long context

A 1M-token window with up to 128K tokens of output. On ProgramBench, a long-context coding test, it scored 79.7%.

Computer use and vision

80.1% on OSWorld 2.1 and 61.6% on Chartography. The first Sonnet model to beat Pokémon Red from screenshots alone.

Specifications

What Anthropic has published for Claude Sonnet 5.5.

Pricing

Claude Sonnet 5.5: $2 / $10 per million tokens, with cache reads at $0.20 and cache writes at $2.50.

Standard rates. Prompt caching saves up to 90% on input, batch processing 50%, and US-only inference costs 1.1x.

Context window

claudeSonnet55.specs.contextBody

The same window as Opus 5.5 and Fable 5.1. Context windows are evaluation-dependent and do not exceed 1M tokens.

Knowledge cutoff

June 2026, matching Opus 5.5 and Fable 5.1.

Modalities

Text and images in, text out, with vision, tool use and multilingual support.

Reasoning effort

Low, medium, high, xhigh and max, with high as the API default and medium in the Claude apps.

Data retention

Available with zero data retention. Cyber and biology safeguards fall back to Sonnet 5 for flagged high-risk requests.

Where Sonnet 5.5 runs

Anthropic ships Sonnet 5.5 on its own platform and the major clouds, and it is already in developer tools.

Claude Platform

Native API under the model ID claude-sonnet-5-5.

Amazon Web Services

Available on Amazon Bedrock as anthropic.claude-sonnet-5-5.

Google Cloud

Available on Vertex AI under the same model ID.

Microsoft Foundry

Available on Microsoft Foundry and Microsoft Azure.

GitHub Copilot

Generally available to Pro, Pro+, Max, Business and Enterprise on 28 September 2026.

Felo AI Search and Felo API

Run it in Felo AI Search or call it through the Felo OpenAPI.

Claude Opus 5.5Claude Sonnet 5

Frequently asked questions

Claude Sonnet 5.5 is Anthropic's Sonnet-tier model, released on 28 September 2026 as the second model in the Claude 5.5 family. It is a faster, lower-cost complement to Opus 5.5, built for well-scoped everyday work like fixing bugs, building features and creating polished documents, slides and spreadsheets. It costs $2 per million input tokens and $10 per million output tokens.

Try Claude Sonnet 5.5 on Felo

Anthropic's Sonnet-tier model: 70.6% on Terminal-Bench 4.0 at $2 in and $10 out per million tokens, with a 1M-token window.

Open Claude Sonnet 5.5 Call via Felo API

Available on Felo AI Search and the Felo API