Claude Opus 5.5: What's New, Benchmarks, and Pricing

On September 22, 2026, Anthropic released Claude Opus 5.5, the first model in its new Claude 5.5 family. The short version: it performs at roughly Claude Fable 5.1 level on most work, costs about 40% less to run than Opus 5, and generates output more than 30% faster. It also posts Anthropic's best-ever safety scores. Here's what actually changed, what the benchmarks say, what it costs, and whether it's worth switching your agents over.

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's new flagship model and the opening release of the Claude 5.5 generation. Per Anthropic's announcement, it's a major step up from Opus 5 on the hardest work — long codebase migrations, multi-repo engineering, and computer-use tasks — while using less compute per task.

It's notable for a second reason: it's Anthropic's first release since the company called for "pacing the frontier." The pitch this time is less "raw capability at any price" and more "frontier-level results, cheaper and safer."

Claude Sonnet 5.5 and Haiku 5.5 are slated to follow in the coming weeks with similar improvements.

Claude Opus 5.5 benchmarks

On Anthropic's published table, Opus 5.5 leads in agentic coding, computer use, and knowledge work. All Opus 5.5 figures use adaptive thinking at max effort unless noted.

Source: Anthropic.

Two honest caveats. First, Anthropic itself says that at this level, benchmark margins are a weaker guide to real-world quality — in its own use, the gap between Opus 5.5 and Fable 5.1 is narrower than the scores suggest. Second, some scores are held back by safeguards: on tasks where guardrails intervened, cybersecurity work fell back to Opus 4.8 and biology work to Opus 5, which likely lowered the reported numbers.

The real story: efficiency

Where Opus 5.5's advantage is clearest is cost per unit of work. It costs less per token and uses fewer tokens per task, which nets out to Anthropic's headline 40% drop.

The coding examples are concrete:

• An early tester audited and fixed a 200,000-line codebase in under three hours, where Opus 5 took over 20 hours and used 2.5x as many tokens.
• Another completed a 680,000-line code migration in less than a day — work Anthropic frames as weeks for an engineering team.
• On an internal C-to-Rust rewrite of HAProxy, Opus 5.5 finished in 9.5 hours versus 12 for Fable 5.1, and cost 51% less.

Anthropic also reports it beats GPT-6 Astra on FrontierCode at roughly 20% of the cost per task, and matches Astra on Terminal-Bench 4.0 for about 40% of the cost. For agentic workloads — where you pay for many steps and lots of re-read context — that efficiency compounds fast.

Claude Opus 5.5 pricing

Opus 5.5 is priced below Opus 5 across the board:

The cache read cut is the one that matters most for agents: at $0.20 per million (60% cheaper), it directly lowers the cost of the re-read-heavy context that dominates coding and multi-step agent bills.

There's also a Fast mode in Claude Code and the Claude Platform running up to 2.5x faster, at $8 input / $40 output per million tokens. On top of the price drop, Anthropic raised five-hour usage limits on Pro, Max, and Team plans and is giving subscribers a rate-limit reset they can save and use when they choose.

Safety and the fallback catch

Anthropic says Opus 5.5 is the strongest-performing model to date on its automated behavioral audit — its most comprehensive alignment test. It's more resistant to prompt injection than Opus 5 and less likely to take hard-to-reverse actions or step outside its given boundaries. It was tested pre-release by external evaluators including METR.

There's one operational detail worth flagging. Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, it ships with Fable 5.1-class safeguards — and on some dual-use tasks those safeguards route the request to an older model (Opus 4.8 or Opus 5) rather than answering directly. As The New Stack noted, that means an agent call can quietly fall back to a different model in the background. If you're building on Opus 5.5, it's worth knowing when that can happen.

Vetted organizations can apply to Anthropic's Life Sciences Verification Program now, with expanded Cyber Verification Program access coming in the following weeks.

Should you switch to Opus 5.5?

A quick read by use case:

• Long-horizon coding agents — likely yes. The token efficiency plus the 60%-cheaper cache reads are aimed squarely at multi-step, context-heavy runs.
• High-volume, cost-sensitive workloads — yes; the 40% cost drop is the headline reason to move Opus 5 traffic over. Test on your real workload, since savings depend on your cache-read share.
• Latency-sensitive apps — the 30%-faster output (or Fast mode) is a real gain over Opus 5.
• Dual-use cyber/bio work — expect safeguard routing to older models on some tasks; plan around the fallback rather than assuming Opus 5.5 answers everything.

For most builders, the summary is simple: near-Fable-5.1 quality, clearly cheaper and faster on the workloads that used to be expensive, with the best safety scores Anthropic has shipped.

Do this with your own AI workforce

Opus 5.5's biggest gains show up on long-running, tool-heavy jobs — codebase-wide migrations, audits, multi-repo engineering — which is exactly the shape of real agent work. Eigent is an open-source "Cowork" desktop app that runs a multi-agent AI workforce locally, so you can point frontier models like Claude Opus 5.5 at coding and research workflows on your own machine. See how it handles multi-step engineering in our review GitHub PRs workflow, or download Eigent and wire up your own agents today.