Claude Opus 5.5 is here: Fable-level smarts at a lower price

Anthropic released Claude Opus 5.5 on September 22, 2026. It is the first model in the Claude 5.5 family and the new default Opus. The pitch is simple: you get performance at the level of Claude Fable 5.1, Anthropic's top model, on most tasks, for less money than Opus 5 cost.

What is new in Opus 5.5

Opus 5.5 is built for the same jobs as Opus 5: long-running agentic coding, research and complex knowledge work. The difference is how well and how efficiently it does them.

• Fable-class quality on most work. Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1 on most tasks. On several coding and agent benchmarks it scores higher.
• Fewer steps to finish a job. In early testing, Opus 5.5 regularly solved tasks in far fewer turns than Opus 5. That is where most of the savings come from.
• The same big context. It keeps the 1 million token context window and the 128K token maximum output of Opus 5.
• Fresher knowledge. Its training data runs to June 2026.

Pricing: cheaper per token, and fewer tokens

The per-token price drops by 20%, and cache reads drop by 60%:

So where does "40% cheaper" come from? The lower price is only half of it. Opus 5.5 also tends to use fewer tokens and fewer turns to reach the same answer, so the total bill for a task shrinks more than the price list suggests.

Benchmarks

Here is how Opus 5.5 compares on a selection of the benchmarks Anthropic published:

The coding results stand out. On Terminal-Bench 4.0, Opus 5.5 jumps 14 points over Opus 5 and more than 10 points over Fable 5.1. It is not a clean sweep, though. OpenAI's GPT-6 Astra still leads on AutomationBench and Terminal-Bench-Science.

As always, benchmarks are a guide, not a guarantee. Test the model on your own tasks before you commit.

Safety

Anthropic calls Opus 5.5 its best-behaved model so far on its automated behavioural audit, which covers nearly 2,000 scenarios. According to the announcement, it is:

• much less likely to take hard-to-reverse actions without checking first,
• more resistant to prompt injection than Opus 5, and
• 85% less likely than Opus 5 to try to get around a boundary it has been given.

Opus 5.5 also ships with the stricter biology and cybersecurity safeguards first used on Fable 5.1. Legitimate researchers in those fields can apply for Anthropic's verification programs to reduce false refusals.

Where you can use it

• Claude apps: available on the Pro, Max, Team and seat-based Enterprise plans.
• Claude API: use the model ID claude-opus-5-5.
• Cloud platforms: Amazon Bedrock (anthropic.claude-opus-5-5), Claude Platform on AWS, Google Cloud and Microsoft Foundry.

Opus 5 remains available everywhere, so you can switch at your own pace.

For developers: what changes when you upgrade

Most Opus 5 prompts work well on Opus 5.5 straight away. The API, however, has a few breaking changes. Check these before you swap the model ID.

The default effort is now medium

This one is easy to miss because nothing breaks. Opus 5 ran at high effort by default. Opus 5.5 defaults to medium. For hard coding or agent tasks, set the effort level yourself:

import Anthropic from '@anthropic-ai/sdk';

const client = new Anthropic();

const message = await client.messages.create({
  model: 'claude-opus-5-5',
  max_tokens: 16000,
  // Thinking is always on. Effort controls how deeply the model thinks.
  output_config: { effort: 'high' }, // low | medium | high | xhigh | max
  messages: [{ role: 'user', content: 'Review this pull request for bugs.' }],
});

for (const block of message.content) {
  if (block.type === 'text') console.log(block.text);
}

If speed matters more than depth, for chat or simple lookups, try low instead.

Progress notes have moved

During long agent runs, the short notes the model writes between tool calls ("found the bug, now fixing the test") now arrive as thinking blocks, which are empty by default. If your app shows those notes to users, it may look like it has gone silent. Set thinking: { type: "adaptive", display: "updates" } (currently a beta feature) to get them back as short summaries.

Should you switch?

For most teams, yes. Opus 5.5 is better than Opus 5 on almost every measure and cheaper to run. That combination is rare.

• Using Claude in the apps? On a paid plan you already have it. Pick Opus 5.5 from the model menu.
• Building on the API? Run your test suite against claude-opus-5-5, fix the four breaking changes above, and set your effort level on purpose.
• Paying for Fable 5.1 today? Try Opus 5.5 on your workload. If it holds up, you could cut your bill by more than half, since Fable 5.1 costs $10 / $50 per million tokens.

What comes next

Opus 5.5 is the first of the Claude 5.5 family. Claude Sonnet 5.5 followed on September 28 as the faster, lower-cost option for everyday work, and Anthropic says Claude Haiku 5.5 is on the way. We will cover each one as it lands.