Skip to content
Advertisement
Anthropic’s Opus models show steady progress from 4.5 to 5.5

Anthropic’s Opus models show steady progress from 4.5 to 5.5

Anthropic / Wikimedia Commons (Public domain)

Side-by-side outputs from the same prompt highlight how far Claude Opus has moved in under a year, with lower prices and higher benchmark scores

Give the same prompt to every Claude Opus model from 4.5 through 5.5, and you get what amounts to a progress report on Anthropic’s flagship line.

The comparison has been circulating on X, and it lands at a moment when the spread between those versions is unusually wide. The newest model, Opus 5.5, arrived with a bigger memory, a smaller price tag, and benchmark numbers that make its predecessors look like rough drafts.

From 4.5 to 5.5 in under a year

Opus 4.5 was released on November 24, 2025. It shipped with a 200K context window and pricing of $5 per million input tokens and $25 per million output tokens.

Opus 5.5 followed on September 22, 2026. Its context window jumps to 1M, so the model can hold roughly five times as much material in view at once.

Pricing moved the other way. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, a 20% cut to base rates. Cache reads, which cover text the model has already processed and can reuse, are up to 60% cheaper.

Advertisement

The benchmark gap

The clearest number in the comparison comes from Terminal-Bench 4.0, which tests how well a model handles work in a command-line environment. Opus 5.5 scored 64.8% there. Opus 4.8, a much more recent model than 4.5, scored 23.6%.

That is a gap of more than 40 percentage points between two releases in the same family.

Anthropic is also making a bolder positioning claim. The company says Opus 5.5 performs at the level of Fable 5.1, which previously sat above Opus in its lineup, while costing approximately 40% less to run than Opus 5.

On several benchmarks, Opus 5.5 is described as matching or beating the higher-tier Fable models outright.

Early users have described the practical effect in terms of time saved. Testers reported finishing complex coding migrations in hours rather than weeks, including one migration covering 680,000 lines of code.

Safety scores and the current lineup

Anthropic says Opus 5.5 posted the highest scores to date on its automated behavioral alignment audit. The model was also evaluated externally before its release.

As of mid-November 2026, Anthropic’s API still offers every Opus version from 4.5 through 5.5. The company recommends Opus 5.5 for everyday use, citing its balance of capability, speed, and pricing for complex tasks.

What the version-by-version view tells us

The Opus line has been defined by steady iteration on a few fronts: coding performance, agentic task completion, reasoning consistency, and cost efficiency. Agentic tasks are jobs where the model takes a series of actions on its own, like editing files or running commands, instead of answering one question at a time.

What this means for the AI market

For companies building on top of AI models, the economics are the headline. A 20% cut to base token rates, plus cache reads that can be up to 60% cheaper, lowers the cost of running heavy workloads like code review, data analysis, and document processing.

Anthropic’s pitch with Opus 5.5 is that customers no longer have to choose between capability and cost. The release also complicates Anthropic’s own product ladder. When Opus 5.5 matches or beats Fable models on multiple benchmarks at a lower operating cost, the question becomes what the premium tier is for.

For developers, early claims about hours-long migrations replacing weeks of work will need to hold up across messier, real-world codebases. The availability of every Opus version from 4.5 onward gives teams room to run their own side-by-side tests.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.