Claude Opus 5.5: what it does, what it costs, and when to use it

Anthropic released Claude Opus 5.5 on September 22, 2026, the same day OpenAI shipped GPT-6 Sol and GPT-6 Luna. The short version: Opus 5.5 is Anthropic’s new flagship, it performs at the level of Claude Fable 5.1 on most work, and it costs about 40% less to run than Opus 5 on typical workloads. Anthropic’s launch post says as much in its summary line, and the rest of this piece unpacks what that means when you actually put the model to work.

What is Claude Opus 5.5?

Opus 5.5 is the first model in Anthropic’s new Claude 5.5 family. Like the rest of the Opus tier, it is a hybrid reasoning model built for serious coding and AI agents, with a 1M token context window. It supports adaptive thinking, so you can run it at different effort levels, and Anthropic’s benchmark notes reference runs at max and xhigh effort.

Where does it fit in the lineup? Anthropic places it at the top for agentic coding, computer use, and knowledge work. The launch post frames it as the model for long-running, highly capable agents. One early tester quoted in the post completed a 680,000-line code migration in under a day, work the post says would have taken an engineering team weeks. Another asked several Claude models to build a game from a single prompt, and Opus 5.5 scored highest on graphics and polish.

Those are vendor-published early results, not independent benchmarks. Treat them as a signal of intent: this model is aimed at big, multi-step work.

How much does Opus 5.5 cost?

List prices, per 1M tokens:

Item	Price
Input	$4
Output	$20
Cache write	$5
Cache read	$0.20
Fast mode input / output	$8 / $40
US-only inference	1.1x input and output

Two things stand out. Input and output are 20% below Opus 5, and cache reads are 60% below, at $0.20 per million. Anthropic says cached reads make up the majority of agentic and coding costs, so that cache price is where a lot of the savings live for long sessions and agents.

The “40% less than Opus 5” claim has two parts: the lower list price, and fewer tokens per task. Anthropic’s page says the model uses fewer tokens to finish the same work, and third-party testers quoted in the launch post, including people running GitHub Copilot and Lovable, reported the same efficiency pattern. So the savings compound rather than just tracking the price cut.

Fast mode is the option to know about if latency matters. It costs double, $8 and $40 per million input and output tokens, and Anthropic lists up to 2.5x faster generation.

What is Opus 5.5 best at?

Anthropic’s product page breaks the strengths into a few clusters:

• Agentic coding: building features, debugging, refactoring, and code review in large codebases, including finding the root cause before changing anything.
• Long-running agents: orchestrating multi-tool tasks, coordinating subagents, and using memory to learn across sessions with minimal oversight.
• Computer use and vision: reading dense documents, charts, and screenshots, and handling multi-step tasks that span multiple applications.
• Knowledge work: professional documents, spreadsheets, and research, with an emphasis on following instructions precisely and staying in scope.

Anthropic also calls out the writing. Early testers found the output clearer and better structured, and the post quotes one saying “it writes the way I do.” If your product surfaces model output directly to end users, that kind of change is worth testing for yourself, since tone is hard to judge from a spec sheet.

On safety, Anthropic says Opus 5.5 posts the best score to date on its automated behavioral audit, is more resistant to prompt injection than Opus 5, and ships with the same class of safeguards as Fable 5.1 in cybersecurity, biology, and anti-distillation.

Should a team on Opus 5 upgrade?

The upgrade math is unusually one-sided this time. Opus 5 lists at $5 and $25 per million input and output tokens, with cache reads at $0.50. Opus 5.5 is $4 and $20 with cache reads at $0.20, generates output more than 30% faster, and uses fewer tokens per task. You get a better model and a lower bill at the same time, which is not how flagship releases usually go.

The API model id is claude-opus-5-5, so the migration is a one-line change for most integrations. The reasons to stay on Opus 5 are narrow: a platform you depend on has not added Opus 5.5 yet, or you have a pinned evaluation that has not been rerun.

If you are choosing inside the Anthropic family rather than upgrading within it, the live comparison is Opus 5.5 versus Sonnet 5 at half the list price. That comparison has its own tradeoffs, covered in Opus 5.5 vs Sonnet 5.

What does this mean for consumer product teams?

If you run a consumer app or website with AI features, a flagship model is a cost of goods decision as much as an engineering one. Every active user who triggers the model spends tokens, and at $4 and $20 per million, heavy users add up faster than light ones.

Two things to check before moving a user-facing feature to Opus 5.5:

1. Where the feature is billed. If the AI feature unlocks digital functionality inside a native app, Apple and Google store billing rules still apply, and a model swap does not change that. The billing model for in-app AI features is covered in app monetization.
2. What the swap does to your numbers. A model change moves quality, latency, and cost at once. If retention or support volume moves the same week, you want to know which change caused it.

If the cross-vendor question is on your list, Opus 5.5 vs GPT-6 Sol compares the two same-day releases directly.

Sources

• Introducing Claude Opus 5.5 (Anthropic)
• Claude Opus (Anthropic)
• Claude Sonnet (Anthropic)
• GPT-6 Sol model docs (OpenAI), for the same-day release context

FAQ

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's flagship model, the first release in the Claude 5.5 family, announced September 22, 2026. Anthropic says it performs at the level of Claude Fable 5.1 on most work.

How much does Claude Opus 5.5 cost?

The API lists $4 per million input tokens, $20 per million output tokens, and $0.20 per million cached input tokens. Fast mode costs $8 and $40 per million input and output tokens. Anthropic estimates typical workloads cost about 40% less than on Opus 5.

Should I upgrade from Opus 5 to Opus 5.5?

For new work, yes. Opus 5.5 is cheaper per token, generates output more than 30% faster, and uses fewer tokens per task. Opus 5 still works if a platform has not added Opus 5.5 yet.

Is Opus 5.5 better than Sonnet 5?

They are built for different jobs. Opus 5.5 is Anthropic's strongest model for hard agentic coding, long-running agents, and computer use. Sonnet 5 is positioned for real-time agents and high-volume work at half the list price.