Published

Anthropic Releases Claude Opus 5.5 With 40% Lower Costs and Faster Performance
Image: ZDNET

Technology and Science · updated 51m ago · 2 min read

Anthropic Releases Claude Opus 5.5 With 40% Lower Costs and Faster Performance

Happened

Opus 5.5 matches Fable 5.1 performance with about 40% lower running cost. Available on AWS, Google Cloud, and Microsoft Azure for enterprise use.

Split on

Quartz.

Left out

6 of 7 outlets skipped it: opus 5.5 needs fewer tokens and answers are 40% less verbose.

14outlets compared

ADSLZoneCadena 3 ArgentinaCriptotendenciasDiarioBitcoinMacRumorsmadboxpcMarkTechPostMashable

Same story, two versions

tap a side to read it in full

The DecoderThe Decoder

promises less "Claudish" writing
Read the original

TechCrunchTechCrunch

changes to how Opus communicates
Read the original
VS

It steers attention from engineering trade-offs to conversational style.

Opus 5.5 launches

Anthropic released Claude Opus 5.5 on September 22, 2026, positioning it as the first model in a new Claude 5.5 family and saying it is its most advanced model to date. Anthropic said Opus 5.5 is 40% cheaper to run on typical workloads than Opus 5, and it set the price at $4 per million input tokens and $20 per million output tokens. Anthropic also said Opus 5.5 generates output more than 30% faster than Opus 5, and it described the model as capable of tackling complex, long-running tasks that analyze information, use tools, or modify large amounts of code.

Anthropic said one evaluator completed a migration of 680,000 lines of code with Opus 5.5 in less than a day, while the company said the same job would have taken weeks for an engineering team. Anthropic said it strengthened security measures to prevent the model from making unauthorized decisions or falling prey to certain attacks, and it said Opus 5.5 is designed to be safer for tasks involving cybersecurity and biology.

Image from ADSLZone
ADSLZoneADSLZone

Coding, speed, and cost

Anthropic said Opus 5.5 is oriented toward full code migrations and audits of large projects, and it cited an internal test where the model audited and corrected a codebase of 200,000 lines in less than three hours. Anthropic said Opus 5.5 reached 66.4% in Terminal-Bench 4.0 and 54.4% in FrontierCode v1.1, while the company also claimed it achieved performance comparable to Claude Fable 5.1 across tasks. Anthropic said cache reads now cost $0.20 per million tokens, 60% less than before, and it estimated that typical workloads could see total cost reduced by about 40%.

Anthropic said output generation is more than 30% faster than Opus 5, and it offered a fast mode in Claude Code and Claude Platform that can multiply speed by 2.5 at a higher rate. Anthropic said Opus 5.5 is available through Amazon Web Services, Google Cloud, and Microsoft Azure, and it said the model is accessible via the Claude Platform as claude-opus-5-5.

Image from Cadena 3 Argentina
Cadena 3 ArgentinaCadena 3 Argentina

Safety and next models

Anthropic said Opus 5.5 is subject to safeguards similar to those on Fable 5.1, and it said those safeguards limit use for discovering exploits in compiled programs or for developing recognizable biological weapons. Anthropic said most cybersecurity tasks will be rerouted to Opus 4.8, and it said vetted organizations can apply to the Life Sciences Verification Program for broader access. CEO Dario Amodei wrote that "I have come to the conclusion that fully addressing the risks requires even more prudence," and Anthropic said it is pacing capability advancement so risk prevention can keep up.

Anthropic said it is preparing more advanced training and evaluation systems for future models, including improvements in security and monitoring, and it said it expects to share more details soon. Anthropic said it is also preparing Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks, extending the Claude 5.5 family beyond Opus 5.5.