Claude Opus 5.5: Top-Tier AI at 40% Less

Anthropic released Claude Opus 5.5 on September 22, another step toward AI that can handle longer, more complex work with less supervision.

The headline isn't the benchmark scores. It's the price-to-capability ratio:

• Top-tier performance. Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1, its most capable public model, on most work: coding, research, professional analysis and long-running AI agents.
• About 40% cheaper to run than Opus 5. Prices dropped 20%, to $4 per million input tokens and $20 per million output. The rest of the savings comes from the model using fewer tokens to finish the same task.
• Cheaper memory for agents. Cache reads, which Anthropic says make up most of the cost of agent and coding workloads, dropped 60% to $0.20 per million tokens.
• 30% faster output than Opus 5.

For context, Claude Fable 5 launched in June at $10/$50 per million tokens. Three months later, comparable capability costs less than half that.

Safer to leave running

The less obvious improvement matters just as much for businesses. Anthropic says Opus 5.5 is its strongest-performing model yet on its automated behavioral audit, and that it is "much less likely than recent models to take hard-to-reverse actions or act outside the boundaries it's been given."

That's the property you want before handing an AI agent access to your inbox, your files or your billing system.

Why this matters for small businesses

Tasks that used to make sense only as occasional AI requests become practical to run every day when the strongest models get cheaper and more reliable at the same time:

• Analyzing large amounts of documents or data
• Maintaining and updating software
• Running multi-step workflows that act on their own

The frontier of AI isn't advancing only through smarter models anymore. Capability, autonomy and cost are improving together. Pair a model like Opus 5.5 for the hard steps with cheap decision models like Jev for the routine ones, and workflows that were too expensive a year ago start to pencil out.

What to do

1. Revisit the AI projects you shelved on cost. A 40% drop in running cost can flip the math.
2. If you run Opus 5, test Opus 5.5 on your real workload. The savings come partly from using fewer tokens, so measure your actual bill, not the price sheet.
3. Start agents with narrow permissions. Better guardrails in the model don't replace limits you set yourself.

---

Wondering which of your workflows just became worth automating? Let's talk.