Claude Opus 5.5: Cheaper and faster than Opus 5

What's new

Anthropic announced Claude Opus 5.5 on September 22, 2026. The model is available through Claude Platform under the name claude-opus-5-5, as well as on AWS, Google Cloud and Microsoft Azure. Anthropic's announcement does not say whether anything special applies to Denmark or the EU, and a monthly price for regular users has not been disclosed.

According to Anthropic, the model costs 40 percent less to run than Opus 5 and generates output 30 percent faster.

What's smart

The two big advantages are speed and price. Faster responses mean shorter wait times when the model writes text or code. Lower operating costs make it cheaper to use at scale.

Anthropic also highlights an example from a test user who completed a code migration of 680,000 lines in under a day. A migration is the move of a large program to a new technology or setup. This is an example from the company itself.

In Anthropic's own tests, the model scores 54.4 percent on FrontierCode v1.1 versus 50.3 percent for Claude Fable 5.1. On GDPval-AA v2.1, which measures work tasks, it gets 1,846 Elo versus 1,735 for Fable 5.1. Elo is a scoring system known from chess, where higher numbers are better. The figures come from the manufacturer.

Cheaper or better?

Cheaper: The API price is $4 per million input tokens and $20 per million output tokens. Tokens are the pieces of text the model reads and writes. That is 20 percent below Opus 5. When the model reuses text it has already seen (cache), it costs $0.20 per million tokens, 60 percent below Opus 5.

Better: According to a blog post from the service OrcaRouter, Opus 5.5 ranks first on the Epoch Capabilities Index with 167.35 points as of October 2, 2026. The index combines more than 50 tests. Right behind it come GPT-6 Astra with 166.51, Claude Sonnet 5.5 with 165.2 and Claude Fable 5.1 with 164.82. OrcaRouter itself assesses that the gap to second place is not statistically meaningful, because the uncertainty intervals overlap heavily. The ranking is based on OrcaRouter's post and has not been confirmed by Epoch AI.

Against the competitor GPT-6 Astra, price is a clear difference. According to OrcaRouter, it costs $10 for input and $50 for output per million tokens, or 2.5 times as much. The open-weights model Kimi K3, whose architecture is publicly available, sits at 157.61, about 10 points behind the leaders.

What it's good for

Anthropic's own examples point toward programming and large, heavy workloads where the model has to work independently for a long time. This includes, for example, updating large codebases.

For companies that build their own solutions on top of Claude, the lower price and higher speed are the most concrete benefits. The same tasks can be done faster and at a lower cost than with Opus 5. For regular users, the benefit depends on how the model becomes available in the apps they use. That has not been disclosed.