Anthropic unveils Claude Opus 4.7, tops GDPval-AA ahead of Claude Mythos

Anthropic has just revealed its latest model, Claude Opus 4.7. It is not as strong as Claude Mythos in cyber security, but it outperforms the other available models on an assessment designed around real-life tasks.

The race between AI labs is relentless, forcing them to ship new models on a tight cadence. This week, Anthropic is rolling out Claude Opus 4.7. While it does not match Claude Mythos for cyber security capability, it represents a major step forward for people using AI to automate office work or write software. Previous Claude models already enjoy a strong reputation among businesses, and Anthropic is adding new strengths to its toolkit with Opus 4.7.

Claude Opus 4.7: a new productivity tool

For coding, Claude Opus 4.7 can independently handle tasks that previously required human oversight. For broader office workflows, the new AI also benefits from improved vision, enabling it to “see” user-provided documents at higher resolution. Opus 4.7 is also better at following instructions, making it easier for users to get the outcomes they are aiming for. Anthropic says this new AI follows instructions literally, and that prompts which were ineffective on earlier models could now produce strong results.

For those relying on AI to generate documents or visuals, Anthropic also claims Claude Opus 4.7 is more creative, improving output quality. Claude Opus 4.7 may also shine in finance, as it beats all other currently available AI systems on a benchmark tailored to that field. Most importantly, Claude Opus 4.7 is now the leader on GDPval-AA, ahead of OpenAI and Google’s models. This evaluation estimates an AI’s ability to handle everyday “real world” tasks, spanning 44 professions across 9 sectors. Put simply, the new model positions itself as the best tool for automating repetitive work.

A test run ahead of Claude Mythos

This April, Anthropic also introduced an ultra-powerful AI model called Claude Mythos. Unlike Opus 4.7-which is already available across the start-up’s products-Claude Mythos is currently restricted to a carefully selected set of organisations. The reason? Mythos already has highly advanced skills in finding security vulnerabilities and could, as a result, trigger a wave of cyberattacks.

Even so, Anthropic intends to make Mythos available to the general public once it can do so safely. In that context, the release of Claude Opus 4.7 lets the company test measures it could later apply to Claude Mythos to stop hackers using the AI to launch cyberattacks. According to Anthropic, the public version of Claude Opus 4.7 includes “protective mechanisms that automatically detect and block requests that indicate prohibited or high-risk uses in cyber security”. “What we learn from deploying these safeguards in practice will help us reach our ultimate goal: wide distribution of Mythos-class models”, the company adds.

Some businesses will still be able to access an unrestrained version of Claude Opus 4.7 for legitimate uses. However, they will need to enrol in a dedicated programme.

What we think

Anthropic has made businesses (and professionals) its core niche. With that in mind, it makes sense that the company focused this new model on the qualities that matter most to that audience. An independent benchmark that aims to measure productivity gains from AI backs up that positioning.

All indications suggest that Anthropic’s main rivals (Google and OpenAI) may head in a similar direction. While broad consumer popularity is a win, businesses and professionals working with a chatbot are typically more willing to pay for a subscription.