AI news · Friday, July 24, 2026

Anthropic releases Opus 5, closing the gap on its Fable model

Anthropic just dropped Opus 5, its newest AI model that sits right below its high-end Fable 5 in power but costs half as much. The company is positioning this as an everyday workhorse, specifically touting its ability to verify its own work and fix complex coding bugs. It’s a bit of a pivot for Anthropic, which has been under the microscope lately regarding its relationship with the U.S. government after its previous models were flagged for potential misuse in cybersecurity. Anthropic claims Opus 5 has stronger safety guardrails and will be less likely to trigger restrictive blocks compared to its predecessors. Interestingly, the company is also adding a new 'automatic fallback' feature for API users, which silently routes requests to a smaller model if the safety system blocks a prompt, ensuring the app doesn't just crash. It’s a direct attempt to smooth over user frustrations while keeping the model on a tight leash.

Meanwhile, a group of tech giants including Meta, Microsoft, and Nvidia signed an open letter warning Washington against cracking down on open-weight AI models. The letter follows recent heat from the Trump administration, which accused a Chinese firm of 'distilling'—tech-speak for training one model on the output of another—Anthropic’s Fable 5 to build its own powerful model. Industry leaders argue that banning these open-weight models would kill the innovation ecosystem and hurt domestic startups that need them for defensive security research. While the letter didn't name China directly, it’s clearly an effort to keep the borders open for code, even as political tension over IP theft boils over. OpenAI and Anthropic—the current holders of the most powerful 'closed' models—notably did not sign the letter.

Finally, the 'AI-as-a-cost-sink' reality is hitting harder than expected. The U.S. Army recently burned through a massive allocation of usage tokens for its AI assistant, Ask Sage, forcing the service to start clamping down on access after promising 'unlimited' usage. It’s a microcosm of a wider trend where companies are realizing that running these agents 24/7 is not a free lunch. Even Nothing, the phone company, is reorganizing its staff and consolidating markets, which it claims is for efficiency rather than a full exit, while also launching a new AI-native business unit. Everyone is trying to pivot to AI, but the bill is becoming impossible to ignore.

The quick hits

Anthropic launched Opus 5 to compete with its own Fable 5 model at half the price — a move designed to make high-end AI capabilities more accessible for everyday work.
Major tech firms including Meta, Microsoft, and Nvidia publicly pushed back against potential U.S. bans on open-weight AI models — arguing that restrictions would stifle innovation and weaken defensive cybersecurity.
The U.S. Army had to pull back on its 'unlimited' AI access after burning through tokens at a pace that proved unsustainable — a sign that AI costs are reality-checking even the military.

Anthropic releases Opus 5, closing the gap on its Fable model

The quick hits

Sources