← До всіх статей
Anthropic releases Claude Opus 5.5: what changes in pricing, the API and agentic work
Опубліковано 2026-09-285 хв читання
claudeanthropicclaude-opus-5-5apinews
On September 22, 2026, Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 family. According to Anthropic, it performs at the level of Claude Fable 5.1 on most work and, at default settings, costs 40% less than Opus 5 on typical workloads. API pricing is $4 per million input tokens, $20 per million output tokens and $0.20 per million cache-read tokens. The model ID is claude-opus-5-5, and it is available on all platforms, including AWS, Google Cloud and Microsoft Azure (Anthropic announcement).
Your first Python request to claude-opus-5-5, the Messages API format, parameters and streaming are covered in Anthropic API Fundamentals. Agentic coding in the terminal is covered in Claude Code & Agentic Development. The first 2 lessons of every course are free, and full access comes with a 7-day free trial.
The biggest change is the cost of agentic work. Input and output tokens are 20% cheaper, cache reads are 60% cheaper, and Anthropic says cache reads make up the majority of agentic and coding costs.
| Item | Opus 5.5 | Opus 5 | Change |
|---|---|---|---|
| Input tokens | $4 | $5 | −20% |
| Output tokens | $20 | $25 | −20% |
| Cache reads | $0.20 | $0.50 | −60% |
| Cache writes (5 min) | $5 | $6.25 | −20% |
Source: Pricing table in the Anthropic announcement.
Anthropic itself notes that benchmark margins have become a less reliable guide to real-world differences, so test models on your own tasks.
Code written for Opus 5 can return a 400 error after you swap the model ID. Anthropic's docs list four changes:
thinking: disabled and a manual budget_tokens return an error; the effort parameter controls depth and defaults to medium.tool_choice set to any or tool) returns an error. auto and none still work.computer_20251124 computer use tool is no longer accepted.Support for claude-opus-5-5 arrived in Python SDK version 1.8.0 on September 22, 2026 (SDK changelog). Update with one command:
pip install -U anthropic
See the lessons First Steps: Installation and Your First Request and Prompt Caching.
For a team, Opus 5.5 makes long agentic runs in Claude Code and multi-app automations noticeably cheaper. On Gray Swan's benchmark it ties Fable 5.1 for the lowest prompt injection success rate of any model tested, according to Anthropic.
A practical migration order:
medium and compare several levels on your own tasks, as the Opus 5.5 prompting guide recommends;max_tokens: thinking counts toward it even when the thinking text isn't returned;On September 28, Claude Sonnet 5.5 followed: model ID claude-sonnet-5-5, $2 per million input and $10 per million output tokens (Claude Platform release notes, model page). Anthropic said in the Opus 5.5 announcement that Claude Haiku 5.5 will follow in the coming weeks. Compare Claude with ChatGPT, Gemini and other models by task in Model comparison.
$4 per million input tokens, $20 per million output tokens, $0.20 per million cache-read tokens and $5 per million 5-minute cache-write tokens. Fast mode costs $8 and $40 per million input and output tokens.
Update the SDK with pip install -U anthropic and set model="claude-opus-5-5" in your request. Remove thinking: disabled and forced tool_choice, or the API will return a 400 error.
Thinking in Opus 5.5 is always on and adaptive: the model decides how much to think. To get faster, cheaper answers, lower the effort parameter, which defaults to medium.
On all platforms, including the Claude Platform, AWS, Google Cloud and Microsoft Azure. Fast mode is available in Claude Code and the Claude Platform.
Build your first agent on Opus 5.5 during the trial: the 10 lessons of Claude Code & Agentic Development take you from setup to MCP and git workflows. Start 7 days free, then $9 a month or $79 a year, cancel anytime.