CoreCodeBlog

Models

Claude Opus 5.5: pricing, specs and how to use it in Claude Code

Published · Last checked

Claude Opus 5.5 was released on September 22, 2026 for long-running agentic coding and knowledge work. Anthropic recommends starting with it when you are unsure which model to use, and it is Claude Code's default model on the Anthropic API. This post collects its official specs and prices, how it differs from its neighbours, and how to use it in Claude Code. Specs and prices come from Anthropic's Opus 5.5 model page and pricing page, checked on November 2, 2026.

Key points
  • The model ID is claude-opus-5-5, with a 1M-token context window, up to 128K output tokens, and a June 2026 knowledge cutoff.
  • It costs $4 per million input tokens and $20 per million output tokens, 20% less than Opus 5. Cache reads cost 0.05 times the input price, $0.20.
  • Thinking is always on and cannot be disabled; the default effort is medium on the API and in Claude Code.
  • Moving from Opus 5 has four breaking changes, and text between tool calls now comes back in thinking blocks.
  • In Claude Code, pin it with /model claude-opus-5-5; opusplan plans with Opus and executes with Sonnet.

Official specs

ItemClaude Opus 5.5
Model IDclaude-opus-5-5 (anthropic.claude-opus-5-5 on Bedrock)
ReleasedSeptember 22, 2026; retires no sooner than September 22, 2027
Context window1M tokens
Max output128K tokens (300K on the Batches API with a beta header)
ThinkingAdaptive, always on
Default effortmedium
Comparative latencyModerate
Input → outputText and images → text
Reliable knowledge cutoffJune 2026

Pricing

ItemPer million tokens
Input$4
5-minute cache write$5
1-hour cache write$8
Cache read$0.20 (input × 0.05)
Output$20
Batch API50% off input and output: $2 / $10
Fast mode$8 input, $40 output; cache multipliers stack on top; not available with the Batch API

Opus 5.5 costs twice Sonnet 5.5 per token and 40% of Fable 5.1. Both Opus 5.5 and Sonnet 5.5 read the cache at 0.05 times the input price, half the 0.1 most models use, which matters most when, as in Claude Code, most tokens come from the cache. Every model's prices are in Claude API pricing.

Compared with Opus 5, Sonnet 5.5 and Fable 5.1

Opus 5.5Opus 5Sonnet 5.5Fable 5.1
Input / output$4 / $20$5 / $25$2 / $10$10 / $50
Cache read$0.20$0.50$0.10$0.25
Comparative latencyModerate—FastSlower
Claude Code default effortmediumhighmediumhigh

Code that calls the API directly and moves from Opus 5 has four breaking changes (What's new in Claude Opus 5.5):

One more change alters the response without failing any request: text between tool calls comes back in thinking blocks, empty at the default display setting. An application that shows that text as progress has to set display to get it back. In Claude Code, Claude Code handles all of this.

When to use Opus 5.5

Anthropic's models overview recommends starting most work on Opus 5.5. In Claude Code a common split is Opus 5.5 for complex reasoning, multi-file changes and architecture decisions; Sonnet 5.5, at half the price, for everyday coding; and Fable 5.1 for problems Opus still cannot solve at higher effort.

opusplan gets both: Opus reasons and designs in plan mode, then Claude Code switches to Sonnet to write the code. Model and effort advice is in Choosing a Claude model.

Using it in Claude Code

On the Anthropic API, Claude Code's opus alias resolves to Opus 5.5. Aliases move with new releases, so to pin the version, or when using a relay, name the model in full:

/model claude-opus-5-5

For one session, launch with claude --model claude-opus-5-5. To make it the default, set it in ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://corecode.live",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-key"
  },
  "model": "claude-opus-5-5"
}

Opus 5.5 has a native 1M context window, so opus[1m] adds nothing. Claude Code runs it at medium effort; use /effort high or above for hard problems. A top-level effortLevel left in an older settings file does not apply to Opus 5.5, so set the level again with /effort.

Safety classifiers and automatic fallback

Opus 5.5 runs with safety classifiers that most often flag cybersecurity and biology content. When one flags a request, Claude Code re-runs it on another model: biology on Opus 5, cybersecurity on Opus 4.8. CoreCode also serves Opus 5 and Opus 4.8, so the fallback runs as it does on the Anthropic API, and the re-run request is billed at the fallback model's price.

Using it through CoreCode

Both of CoreCode's Claude groups serve claude-opus-5-5, answered by official Anthropic Claude accounts, with each session kept on one account so the cache keeps hitting. The groups are compared in Groups, prices are in the price table, and setup is in Getting started.

FAQ

What does Opus 5.5 improve over Opus 5?

It is 20% cheaper ($4 / $20 against $5 / $25), cache reads drop from 0.1 to 0.05 times the input price, and Anthropic positions it for long-running agentic coding and knowledge work. In exchange, thinking cannot be turned off, and a few API behaviours changed.

Opus 5.5 or Sonnet 5.5?

Sonnet 5.5 costs half as much per token and is faster, which suits everyday coding; use Opus 5.5 for complex reasoning and large changes, or opusplan to split the work.

Can I turn off Opus 5.5's thinking?

No. Its adaptive thinking is always on; effort controls how deep it goes.

How do I confirm I am really getting Opus 5.5?

Run /status in Claude Code for the current model; CoreCode's usage log records each request's model and price. To check that a relay's upstream is Anthropic, see How to verify a Claude relay.

Sources