Anthropic has unveiled Opus 5.5, its newest flagship AI model, with the company touting stronger performance across coding, knowledge work and computer-use tasks at a significantly lower cost than its predecessor. The launch gives Claude AI a new high-end model just two months after the release of Opus 5, while Anthropic says Opus 5.5 delivers performance comparable to its more expensive Fable 5.1 model on most work.
The release arrives at a particularly notable moment for Anthropic, whose CEO Dario Amodei recently called for the AI industry to “pace the frontier” as developers work to keep safety measures aligned with increasingly capable systems. Opus 5.5 is the company’s first model release since that call, and it introduces both performance improvements and additional safeguards aimed at managing the risks associated with more capable AI systems.
Anthropic Says Opus 5.5 Leads On Coding Benchmarks
View this post on Instagram
Announced Tuesday, September 22, Opus 5.5 is the first model in Anthropic’s new Claude 5.5 family. The company says the model delivers its strongest results yet across several categories, including agentic coding, computer use and knowledge work.
On GDPval-AA v2.1, a benchmark measuring AI performance across 44 occupations, Anthropic reported that Opus 5.5 scored 1,846, compared with 1,735 for Fable 5.1 and 1,708 for Opus 5.
The company also reported a 66.4 percent score for Opus 5.5 on Terminal-Bench 4.0, a benchmark focused on agentic software development. Fable 5.1 scored 55.8 percent, while OpenAI’s GPT-6 Astra recorded 57.9 percent, according to Anthropic’s published results.
Opus 5.5 did not lead every benchmark, however. On Terminal-Bench-Science 0.1, GPT-6 Astra scored 64.6 percent compared with 58.7 percent for Opus 5.5. Anthropic has also cautioned that benchmark differences at this level of capability may not directly translate into noticeable differences in everyday use.
The New Claude AI Model Costs Less To Run
Opus 5.5 is here! 🎉 It’s honestly such a nice model to work with.
Feels like Fable, but ~30% faster and ~40% cheaper per task than Opus 5. Go try it out! pic.twitter.com/zwo83sbgez
— Lydia Hallie ✨ (@lydiahallie) September 22, 2026
Price is one of the biggest changes accompanying the launch. Anthropic is charging $4 per million input tokens and $20 per million output tokens for Opus 5.5, compared with $5 and $25 respectively for Opus 5.
The company says the reduction goes beyond the lower token prices. Because Opus 5.5 can complete certain tasks with fewer tokens and generate output more than 30 percent faster, Anthropic estimates that typical workloads can cost about 40 percent less than they did with Opus 5.
Cache reads have also dropped from $0.50 to $0.20 per million tokens, while cache writes have fallen from $6.25 to $5. Anthropic is also introducing a faster mode for Claude Code and its Claude Platform that can run up to 2.5 times faster, although it comes with higher pricing.
The lower costs could be particularly relevant for developers building AI agents, where a single task can involve numerous model calls. Anthropic’s internal and customer testing suggests that Opus 5.5 can sometimes complete complex tasks in fewer steps than Opus 5.
Anthropic Highlights Efficiency Gains
🚨HOLY FUCK! Claude Opus 5.5 Benchmarks is Terrifying
it absolutely destroys Fable 5.1 and GPT-6 Astra and the previous Opus 5 on these benchmarks
66.4% on Terminal-Bench
57.8% on CursorBench
54.4% on Frontier Code
1846 on GDPval-AA
67.7% on Humanity’s Last Exam
81.8% on… pic.twitter.com/QCQPk1JHv1— Salio (@Mr_Salio) September 22, 2026
Several early evaluations cited by Anthropic point to improvements in how efficiently the model handles large technical tasks. Box reported that Opus 5.5 used roughly one-third as many tokens as Opus 5 in its evaluations while producing answers that were shorter without losing accuracy.
GitHub also tested the model in Copilot CLI and VS Code, with Anthropic reporting that Opus 5.5 completed more terminal tasks than Opus 5 while requiring fewer steps. Deloitte, meanwhile, reported improvements in code-review performance at the model’s lowest effort setting compared with Opus 5 at high effort.
These evaluations are not standardized independent benchmarks, and developers will still need to test the model against their own workloads. Still, they form part of Anthropic’s case that the latest Claude AI model is not simply more capable but also more efficient.
Safety Remains A Major Focus
As our models grow more powerful, safeguards are one way we prevent misuse. Opus 5.5 is the first Opus with Fable 5.1-class safeguards on cyber, bio, and frontier LLM development.
Flagged requests fall back to another model, and we’re working to reduce incorrect flags.
— ClaudeDevs (@ClaudeDevs) September 22, 2026
The launch also comes with heightened attention to AI safety. Anthropic says Opus 5.5 was evaluated before release by independent AI safety research organizations Frontier Design and METR.
According to Anthropic, the model performed better than previous systems on its internal evaluations and was about 85 percent less likely than Opus 5 or Mythos 5.1 to attempt to bypass containment boundaries in one dedicated test.
Opus 5.5 also uses safety classifiers covering areas such as cybersecurity, biology and frontier AI development. When a request triggers one of those safeguards, Anthropic says it can be routed to an older model instead.
The company is expanding access to the model for vetted cybersecurity and life-sciences researchers through specialized verification programs, while maintaining additional restrictions around sensitive capabilities.
Sonnet 5.5 And Haiku 5.5 Are Coming
Opus 5.5 is the first release in the Claude 5.5 generation, but it will not be the last. Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will launch in the coming weeks, bringing the same broader focus on performance, speed and safety to the other tiers of its model lineup.
For developers, Opus 5.5 is available through Anthropic as well as cloud platforms including Amazon Web Services, Google Cloud and Microsoft Azure. The launch marks another step in the company’s effort to make increasingly capable AI models more efficient to deploy while expanding the safeguards surrounding their use.
Featured image: Anthropic
—Read Also
