Anthropic launched Claude Opus 5.5 on September 23, 2026 – a cheaper, faster version of its flagship model that the company says matches its more powerful Claude Fable 5.1 on most everyday tasks. It costs 40% less to run than Opus 5 and is available now to Pro, Max, Team, and Enterprise users.
That’s the headline. Here’s what actually changed, what it costs, and whether it’s worth switching your workflow to.
What Is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic’s latest Opus-tier model, aimed at knowledge work, coding, and computer-use agents – the stuff you’d normally reach for the “big” model to do. Anthropic’s pitch is unusual for an AI release: instead of leading with raw intelligence gains, they’re leading with cost. Opus 5.5 is priced 20% below Opus 5, and because it apparently uses fewer tokens to get to an answer, Anthropic is claiming a 40% real-world cost drop on typical workloads.
If that framing sounds familiar, it’s because it’s the same move Opus 5 made against Fable 5 back in July – “almost as smart, way cheaper” is becoming Anthropic’s whole Opus-tier identity.
Claude Opus 5.5 Pricing
Opus 5.5 runs $4 per million input tokens and $20 per million output tokens. Cache reads – which Anthropic says make up most of the cost on agentic and coding workloads – are priced at $0.20 per million tokens, a 60% cut from before.
For context, Opus 5 launched at $5/$25 per million tokens. So the sticker price dropped 20%, and Anthropic’s argument is that the effective price drops further once you factor in fewer tokens burned per task.
What Changed From Opus 5
Three things, according to Anthropic’s own claims:
- Speed: output generation is more than 30% faster
- Cost: roughly 40% cheaper for typical workloads, once token efficiency is factored in
- Writing style: Anthropic says the model sticks closer to a user’s actual instructions, cuts jargon, and leads with the important information instead of burying it
That last point matters more than it sounds. A model that “writes more naturally” is marketing-speak until you’ve spent a few months fighting a model that pads every answer with hedges and caveats. Whether Opus 5.5 actually fixes that is something you’ll only know by using it – Anthropic’s own description of it isn’t proof.
Benchmark Performance
Anthropic put Opus 5.5 up against CursorBench 4.0, a coding benchmark, where it scored 52.5% at default effort. The company’s comparison point was GPT-5.6 Sol, which topped out at 41.7%. Anthropic also claims Opus 5.5 costs about a third as much per completed task as that comparison.
Two things worth flagging: this is Anthropic’s own benchmark run, not an independent one, and CursorBench 4.0 isn’t a standardized industry benchmark the way SWE-bench is. Treat the number as directionally interesting, not gospel, until third-party evals show up.
Safety Claims
Anthropic says Opus 5.5 is about 85% less likely to attempt to escape its containment limits compared to both Opus 5 and Mythos 5.1. Before release, the model went through external review from Frontier Design and METR – though Anthropic notes that review focused on how the model could affect AI R&D, not a full alignment audit.
This release lands about three weeks after OpenAI’s GPT-6 Astra, which Anthropic describes as capable of browsing the web, operating apps, handling spreadsheets and documents, writing and testing its own code, and running multi-step workflows without much hand-holding. The safety framing here reads like it’s partly aimed at that comparison – more autonomous agents, more scrutiny on what happens when they go off the rails.
How to Access Claude Opus 5.5
Opus 5.5 is live now for anyone on a Claude Pro, Max, Team, or Enterprise plan. Developers can reach it through the Claude Platform, AWS, Google Cloud, or Azure – no separate signup required if you’re already building on one of those.
FAQ
Is Claude Opus 5.5 free?
No. It’s included in paid Claude subscriptions (Pro and above) and billed per-token through the API and cloud platforms.
Is Claude Opus 5.5 better than Claude Fable 5.1?
Anthropic says Opus 5.5 matches Fable 5.1 on most tasks, but Fable 5.1 remains the more capable model for the hardest, most demanding work. Opus 5.5 is the cost-efficient option, not a straight replacement.
How much does Claude Opus 5.5 cost per million tokens?
$4 for input tokens and $20 for output tokens, with cache reads at $0.20 per million tokens.
When was Claude Opus 5.5 released?
September 23, 2026.
Our Take
A 20% price cut and a 40% real-world savings claim is a bigger deal than another leaderboard win would’ve been. Most people running Opus in production aren’t bottlenecked on raw intelligence – they’re bottlenecked on cost per task at scale. If the token-efficiency claims hold up outside Anthropic’s own numbers, this is the release that actually changes what’s viable to automate, not just what scores well on a benchmark nobody outside AI Twitter has heard of.
We’ll update this once independent benchmarks and real usage reports start coming in.
