Anthropic has officially released Claude Opus 5.5, the first model in its new Claude 5.5 family. Launched on September 22, 2026, the new model focuses heavily on agentic coding, complex knowledge work, efficiency, and AI safety.
Perhaps the most interesting part of the release is that Anthropic is not simply promising higher benchmark numbers. Claude Opus 5.5 is also significantly cheaper to operate than Claude Opus 5, while Anthropic says it delivers performance comparable to its higher-end Claude Fable 5.1 model on most tasks.
For developers using Claude Code, AI agents, or Claude through an API, this could make Opus-level intelligence much more practical for everyday development work.
What Is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic’s latest Opus-class model and the successor to Claude Opus 5.
Anthropic positions Opus as its model family for complex coding, agentic workflows, reasoning, and professional knowledge work. With Opus 5.5, the company appears to be focusing as much on efficiency as raw intelligence.
According to Anthropic, Opus 5.5 performs at roughly the level of Claude Fable 5.1 for most workloads while costing around 40% less to run than Opus 5 under typical workloads.
That combination is important because powerful reasoning models can become expensive when used inside autonomous agents. A coding agent might repeatedly inspect files, call tools, run commands, analyze results, and revise its implementation before completing a single task.
Reducing the cost of those loops can matter just as much as improving benchmark scores.
Claude Opus 5.5 Pricing
One of the biggest changes is pricing.
Claude Opus 5.5 costs:
- $4 per million input tokens
- $20 per million output tokens
Claude Opus 5 was priced at $5 per million input tokens and $25 per million output tokens, meaning the base token prices have dropped by 20%.
The savings can become even larger for agentic workflows because Anthropic has also reduced prompt-cache read costs by 60%.
This is especially relevant to Claude Code. Long coding sessions repeatedly send large parts of the existing conversation, project instructions, tool definitions, and previously inspected code back to the model.