Anthropic Has Released Claude Opus 5.5
The new model features lower API pricing and improved token efficiency for agentic tasks.
Updated on Sept. 22, 2026 in Artificial Intelligence
Live Poll
Do you trust that lower-cost AI models will perform as reliably as more expensive ones?
Anthropic has released Claude Opus 5.5, a large language model that introduces a 20% reduction in API pricing. Internal evaluations by Box suggest the model requires one-third as many tokens as its predecessor for specific tasks.
Why it matters
This release aims to improve performance in agentic settings where models select their own tools, following Anthropic's implementation of tighter filtering in reinforcement learning environments. It addresses the growing need for cost-efficient, high-reliability AI models that can manage multi-turn workflows.
Claude Opus 5.5 achieved a 66.4% score on Terminal-Bench 4.0 and 54.4% on FrontierCode. The API is priced at $4 per million input tokens and $20 per million output tokens.
The players
Anthropic
An artificial intelligence research company focused on building large language models with a focus on safety and alignment.
Box
A cloud content management and file sharing service that evaluated the token usage efficiency of the new model.
The details
Claude Opus 5.5 uses safety classifiers that can intercept and silently reroute requests to older model versions, such as Opus 4.8 or Opus 5, during multi-turn workflows. This architecture allows the system to prioritize reliability by falling back on proven versions when safety parameters are triggered. By requiring fewer total tokens to complete specific tasks, the model reduces operational overhead compared to prior versions.
Timeline
Anthropic released Claude Opus 5.5 on September 22, 2026.
Sonnet 5.5 and Haiku 5.5 are scheduled for release in the coming weeks.
The Tech Race
Anthropic is extending its current research trajectory by integrating the new model into its broader Cyber Verification Program. This move underscores a competitive push toward high-reliability agentic AI that can be validated for enterprise-grade security environments.
Users can access the new model via API with a 20% price reduction on usage, alongside a 20% increase in subscription usage limits. These changes primarily benefit developers and enterprises building agentic workflows that require high-frequency tool selection.
The takeaway
The trajectory of AI development is shifting toward efficiency and safety-aligned routing rather than simple raw performance gains. Keep an eye on the upcoming release of the Sonnet 5.5 and Haiku 5.5 models to see how these benchmarks hold across different parameter sizes.
What happens next
Anthropic is expected to release Sonnet 5.5 and Haiku 5.5 models in the coming weeks, alongside the integration of Opus 5.5 into its Cyber Verification Program.
Further reading
For broader trends in model development, visit the Artificial Intelligence section.
Live Poll
Do you trust that lower-cost AI models will perform as reliably as more expensive ones?
