Claude Sonnet 5.5 is Anthropic's Sonnet-class model, released on 28 September 2026 as a direct successor to Claude Sonnet 5. It is built for well-scoped everyday work rather than the hardest frontier problems, and it carries a 1,000,000-token context window with always-on adaptive thinking.
Anthropic positions it close to Opus 5.5 on agentic coding and knowledge work while remaining the cheaper, faster tier. On the vendor's own figures it beats Opus 5.5 on Terminal-Bench 4.0 and sits within two points of it on GDPval-AA.
| Developer | Anthropic |
|---|---|
| Model ID | anthropic/claude-sonnet-5.5 |
| Release Date | September 28, 2026 |
| Context Window | 1,000,000 tokens (~1M) |
| Max Output Tokens | 128,000 tokens |
| Modalities | Image, Text → Text |
| Capabilities | reasoning, streaming, tool/function calling, vision |
A large jump in agentic coding
Anthropic reports 70.6% on Terminal-Bench 4.0, against 10.3% for Claude Sonnet 5 on the same evaluation, and 55.5% on CursorBench 4.0 against 34.1%.
Computer use
80.1% on OSWorld 2.1 (partial credit), against 57.0% for Sonnet 5.
Reading charts and screenshots
61.6% on Chartography without tools, against 15.6% for Sonnet 5. Anthropic notes it is the first Sonnet model to finish Pokémon Red working only from screenshots.
Long-context work
The 1M-token window is available at standard pricing on AI/ML API, with prompt caching billed separately.
Agentic coding and long-running tasks
The Terminal-Bench and CursorBench gains over Sonnet 5 are the largest Anthropic reports for this release, and they are the tasks the model is aimed at.
Computer-use agents
80.1% on OSWorld 2.1 puts it within two points of Opus 5.5 at a lower price per token.
Work over very large inputs
A 1M-token window suits whole-repository review, long transcripts and large document sets.
Claude Sonnet 5
The model this one replaces. Sonnet 5.5 is ahead on every benchmark Anthropic published for both, by wide margins on the agentic ones.
Claude Opus 5.5
The frontier tier. Opus leads on CursorBench 4.0 (57.8% vs 55.5%) and FrontierCode, ties on GDPval-AA (1846 vs 1844), and trails on Terminal-Bench 4.0 (66.4% vs 70.6%). Sonnet 5.5 is the cheaper choice where that gap does not matter.
Source: https://www.anthropic.com/claude-sonnet-5-5
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "anthropic/claude-sonnet-5.5", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "anthropic/claude-sonnet-5.5", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"anthropic/claude-sonnet-5.5","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 56 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | September 28, 2026 |
| Terminal-Bench | 70.6% (v4.0) | Autonomous shell/terminal task completion | Source | September 28, 2026 |
| OSWorld | 80.1% (v2.1, partial) | Computer-use across real desktop applications | Source | September 28, 2026 |
| Humanity's Last Exam | 64.5% (with tools) | Expert-level questions across many domains | Source | September 28, 2026 |
Terminal-Bench, OSWorld and Humanity's Last Exam as published by Anthropic at release. Intelligence Index measured independently by Artificial Analysis.
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Claude Sonnet 5.5 This page | Multimodal reasoning | |||
| Complex agentic coding and enterprise workflows | ||||
| Long-horizon agentic coding and research | ||||
| Agentic workflows and structured output |
It is built for coding, agents, and professional work at scale. It is also described as suitable for reasoning, streaming, tools, and vision.
Anthropic positions it for coding, multi-step workflows and autonomous agents, and reports its largest gains over Claude Sonnet 5 on agentic evaluations such as Terminal-Bench 4.0 and CursorBench 4.0. The 1M-token context window also suits long documents and whole-repository work.
Yes. Anthropic says it is designed to navigate real codebases, land multi-file changes, and carry longer debugging and refactoring tasks through to completion.
Confirm that your workflow needs coding, agentic automation, tool use, or vision. The model is positioned for professional tasks at scale, so it is a strong fit when those capabilities matter most.
