Anthropic introduces Claude Opus 5.5, its new lower-cost AI model for agentic coding, computer use, and more
Claude Opus 5.5 outperforms several competing AI models on certain benchmarks, particularly in coding, computer use and workplace tasks.
- Sep 23, 2026,
- Updated Sep 23, 2026 3:00 PM IST
Anthropic has introduced a new low-cost AI model, the Claude Opus 5.5, that offers similar capabilities and safety upgrades as the Claude Fable 5.1 model. The company says the new Opus 5.5 leads in agentic tasks, coding, computer use, and knowledge work, all at 40% less cost compared to Opus 5, which launched in July. Here’s everything you need to know about the new Anthropic model.
Must read: OpenAI expands GPT-6 lineup with Sol and Luna models, bringing cheaper, more efficient AI
Anthropic’s Claude Opus 5.5: Performance and benchmark
In a blog post, OpenAI highlighted that early testers found significant improvements across complex workflows. It highlighted that an early tester completed a 680,000-line code migration in less than a day, which could’ve taken an engineering team several weeks. It can also identify and fix software inefficiencies more quickly.
In benchmarks, Claude Opus 5.5 outperforms several competing AI models on certain benchmarks, particularly in coding, computer use, and workplace tasks. In GDPval-AA v2 testing, which evaluates AI performance on real-world work, Opus 5.5 achieved 1,846, compared to Claude Fable 5.1’s 1,735.
Opus 5.5 also outperformed GPT-6 Astra on Terminal-Bench 4.0. It scored 66.4%, compared to GPT-6 Astra’s 57.9%, highlighting an 8.5% higher efficiency. However, Astra is better on Terminal-Bench-Science 0.1.
Must read: AI safety or market control? Anthropic, OpenAI, Google SpaceXAI accused of slowing AI progress
In terms of safety, Anthropic highlighted that the model has been tested thoroughly, including by outside evaluators like METR and Frontier Design. “It is much less likely than recent models to take hard-to-reverse actions or act outside the boundaries it’s been given, and it’s more resistant than Opus 5 to prompt injection,” the company wrote in the blog.
The company also said that it's preparing stronger safety and monitoring systems for its future AI models. It's also preparing for stronger protections to prevent potential misuse or harmful behaviour. It is also preparing for new methods to train models safely and evaluate their capabilities and risks before deployment.
“As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe. That capacity takes time to build, and we’ve started to put the infrastructure in place to support it,” the blog reads. “We expect to share more details on these efforts soon.”
Claude Opus 5.5 is available across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. It is also available on Claude Platform for developers to use. Anthropic also highlighted that it will soon be launching Claude Sonnet 5.5 and Claude Haiku 5.5 models.
For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine
Anthropic has introduced a new low-cost AI model, the Claude Opus 5.5, that offers similar capabilities and safety upgrades as the Claude Fable 5.1 model. The company says the new Opus 5.5 leads in agentic tasks, coding, computer use, and knowledge work, all at 40% less cost compared to Opus 5, which launched in July. Here’s everything you need to know about the new Anthropic model.
Must read: OpenAI expands GPT-6 lineup with Sol and Luna models, bringing cheaper, more efficient AI
Anthropic’s Claude Opus 5.5: Performance and benchmark
In a blog post, OpenAI highlighted that early testers found significant improvements across complex workflows. It highlighted that an early tester completed a 680,000-line code migration in less than a day, which could’ve taken an engineering team several weeks. It can also identify and fix software inefficiencies more quickly.
In benchmarks, Claude Opus 5.5 outperforms several competing AI models on certain benchmarks, particularly in coding, computer use, and workplace tasks. In GDPval-AA v2 testing, which evaluates AI performance on real-world work, Opus 5.5 achieved 1,846, compared to Claude Fable 5.1’s 1,735.
Opus 5.5 also outperformed GPT-6 Astra on Terminal-Bench 4.0. It scored 66.4%, compared to GPT-6 Astra’s 57.9%, highlighting an 8.5% higher efficiency. However, Astra is better on Terminal-Bench-Science 0.1.
Must read: AI safety or market control? Anthropic, OpenAI, Google SpaceXAI accused of slowing AI progress
In terms of safety, Anthropic highlighted that the model has been tested thoroughly, including by outside evaluators like METR and Frontier Design. “It is much less likely than recent models to take hard-to-reverse actions or act outside the boundaries it’s been given, and it’s more resistant than Opus 5 to prompt injection,” the company wrote in the blog.
The company also said that it's preparing stronger safety and monitoring systems for its future AI models. It's also preparing for stronger protections to prevent potential misuse or harmful behaviour. It is also preparing for new methods to train models safely and evaluate their capabilities and risks before deployment.
“As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe. That capacity takes time to build, and we’ve started to put the infrastructure in place to support it,” the blog reads. “We expect to share more details on these efforts soon.”
Claude Opus 5.5 is available across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. It is also available on Claude Platform for developers to use. Anthropic also highlighted that it will soon be launching Claude Sonnet 5.5 and Claude Haiku 5.5 models.
For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine