Anthropic, the artificial intelligence company behind the Claude chatbot, released Claude Opus 4.7 on Thursday, an upgrade to its flagship model that the company says delivers meaningful improvements in software engineering, image understanding and the ability to carry out complex tasks with minimal human oversight.
The model is available immediately through Anthropic's own products, its developer tools and through cloud computing platforms run by Amazon, Google and Microsoft.
Pricing remains unchanged from the previous version at $5 per million input tokens and $25 per million output tokens, where tokens are the small chunks of text that AI models process when reading and generating language.
Anthropic said Opus 4.7 outperforms its predecessor, Opus 4.6, on a range of industry benchmarks measuring coding ability, reasoning across multiple subjects and the capacity to use software tools independently.
The company published charts showing the model also surpasses OpenAI's ChatGPT 5.4 and Google's Gemini 3.1 Pro on several of those measures.
However, Anthropic acknowledged the model is not its most powerful system.
That distinction belongs to Claude Mythos, a separate model focused on cybersecurity that the company has restricted to a small group of vetted companies and has not made publicly available, citing safety concerns about its ability to autonomously execute sophisticated cyberattacks.
The release arrives after weeks of complaints from developers who said the previous model, Opus 4.6, had quietly deteriorated in quality, a phenomenon users described online as "AI shrinkflation."
One senior director at AMD, the chipmaker, wrote in a widely shared post that Claude had regressed to the point where it could not be trusted with complex engineering work.
Anthropic denied that any deliberate scaling back had taken place.
Early testing by the technology publication Decrypt found Opus 4.7 produced the strongest results it had seen from any AI model on a standardised game-building task, with the most polished output and the most creative design.
But testers also flagged a significant trade-off: the model consumed far more computing resources than its predecessor, with a single session depleting an entire token allowance for the first time in the publication's testing history.
Anthropic said the model uses an updated system for processing text that can result in the same input requiring up to 35% more tokens depending on the type of content, and that the model also does more internal reasoning at higher difficulty settings, producing more output as a result.
For businesses paying by the token, that means the same price per unit could translate into a noticeably higher bill.
The company has published guidance for developers on managing the increase, including new controls that allow them to set spending caps on longer tasks.
Related reading
- Anthropic spots 'emotion vectors' inside Claude
- Anthropic pushes back on claims of Claude degradation as user complaints mount
- Anthropic launches Claude Managed Agents as enterprise revenue hits $30 billion run rate
Anthropic also introduced a new difficulty setting called "extra high" that sits between its existing "high" and "maximum" options, giving developers finer control over the balance between quality and cost.
The model arrives two months after Opus 4.6, maintaining a release cadence of roughly one major upgrade every eight weeks that Anthropic has sustained since late last year.
The recap
- Anthropic releases Claude Opus 4.7 AI model to market.
- Model labelled a "Token Eating Machine" by Decrypt.
- Decrypt published coverage highlighting concerns over token consumption.