Chat ● Operational Service status →

Claude Sonnet 5.5

⧉

claude-sonnet-5-5

Claude Sonnet 5.5 is the newest model in Anthropic's Sonnet line and the successor to Claude Sonnet 5, which Anthropic calls its best combination of speed and intelligence: output more than 30% faster than Sonnet 5, fewer tokens and tool calls for the same task, and 70.6% on Terminal-Bench 4.0. It has a 1M-token context window, up to 128K output tokens per request and adaptive thinking, with image input, tool calling, prompt caching and streaming. KovAPI serves it through both the OpenAI-compatible API and the native Anthropic Messages API (/anthropic), so Claude Code connects directly; effort control, thinking summaries, structured output and PDF input are available on the native API. Pay as you go; failed calls are never billed.

ChatCodeAgents1M contextNative Anthropic API

Get API Key → API Docs

Service Highlights

Anthropic's newest Sonnet, succeeding Sonnet 5
Output more than 30% faster than Sonnet 5; 70.6% on Terminal-Bench 4.0
1M-token context, up to 128K output per request
Five effort levels on the native API; image and PDF input, tool calling, prompt caching
Priced below Anthropic's list on both ends ($2 / $10)

Measured on real calls

Customer requests over the last 7 days

Success rate98.5%

Pricing

Input$1.5 / 1M tokens

Output$7.5 / 1M tokens

List price~~$2.0 / $10.0 / 1M~~

You save ↓25%

View detailed pricing

Why this model

Anthropic's newest Sonnet · the best mix of speed and intelligence

Highlights

Anthropic's newest Sonnet, succeeding Sonnet 5
Output more than 30% faster than Sonnet 5; 70.6% on Terminal-Bench 4.0
1M-token context, up to 128K output per request
Five effort levels on the native API; image and PDF input, tool calling, prompt caching
Priced below Anthropic's list on both ends ($2 / $10)

Price vs. official list

KovAPIList price Input$1.5$2 Output$7.5~~$10~~

Per 1M tokens, USD · You save ↓25%

Live Test · Playground

Try out Claude Sonnet 5.5 right here (available after login).

Input

Model

Advanced

temperature 1.0 max_tokens 2048

Web search

Memory · multi-turn

Thinking

Conversation

Start a conversationType a message below to begin

About Claude Sonnet 5.5

Claude Sonnet 5.5 is the newest model in Anthropic's Sonnet line and the successor to Claude Sonnet 5, at the same list price ($2 / $10 per million tokens). Anthropic calls it the best combination of speed and intelligence: output more than 30% faster than Sonnet 5, fewer tokens and tool calls for the same task, up to about 30% lower cost per task, and 70.6% on Terminal-Bench 4.0. It has a 1M-token context window and up to 128K output tokens per request; thinking is adaptive, so simple requests are answered directly (short questions came back in 1.5-3 seconds in our tests) and hard ones are reasoned through first. On KovAPI, set model to claude-sonnet-5-5 on the OpenAI-compatible API; the native Anthropic Messages API (/anthropic) lets Claude Code connect directly, with effort, thinking summaries, structured output, PDF input and cache_control working as documented. Pay as you go; failed calls are never billed.

Use cases

Everyday coding

Building features, fixing bugs, refactoring and code review: fast and cheap per call, a natural daily driver for coding tools.

Agents and tool calling

Agent loops that call tools over many steps, using fewer tokens and tool calls for the same task.

Production traffic at scale

Support replies, extraction, summarisation and classification: high-volume, latency-sensitive work.

Documents and spreadsheets

Well-structured documents, slide outlines and tables; use structured output on the native API when you need strict JSON.

How to choose

Choose Claude Sonnet 5.5 for everyday coding, agents and high-volume calls: it is fast and costs half as much as Opus 5.5. Choose Claude Opus 5.5 for the hardest reasoning and long-horizon agent work. Switching is just the model field.

FAQ

How is it different from Claude Sonnet 5?

Sonnet 5.5 is the newer model in the Sonnet line, at the same list price as Sonnet 5. By Anthropic's figures its output is more than 30% faster, it uses fewer tokens and tool calls for the same task, it costs up to about 30% less per task, and it scores 70.6% on Terminal-Bench 4.0. Both have a 1M-token context and up to 128K output tokens per request.

Can I set the effort? Can I turn thinking off?

You can set the effort: on the native API choose output_config.effort from low to max, and add thinking: {"type": "adaptive", "display": "summarized"} to get a summary of the thinking. Thinking is adaptive, so simple requests are usually answered directly and there is no need to switch it off; thinking set to disabled is ignored. On the OpenAI-compatible API the model thinks at its default effort, reasoning_effort has no effect, and no thinking content is returned.

Which parameters work, and what are the limits?

messages, streaming, stop, tools and image input all work. Thinking cannot be turned off (thinking set to disabled is ignored); a forced tool choice (any, or a named tool) is treated as auto; temperature, top_p and top_k are ignored. For strict JSON output use output_config.format on the native API; response_format on the OpenAI-compatible API is not guaranteed to take effect.

How do I send images and PDFs?

Images work on both APIs: image_url on the OpenAI format, an image block on the native API, as base64 or a public link. Send PDFs as base64 document blocks on the native API.

Am I charged for failed calls?

No. Only calls that return normally are settled, at actual usage; upstream errors and timeouts never touch your balance.