18 AI Updates This Week: A Free Tool That Merges ChatGPT and Claude, and the Truth Behind Meta's Claude Code Ban

Rather than settling the two-year debate over whether ChatGPT or Claude is smarter, a free tool called Hermes takes a different approach: merge them. Pick any two AIs inside Hermes -- say ChatGPT and Claude -- and it runs them both on the same task simultaneously, then has one read both answers and merge the best of each into a single, cleaner result. Wherever one model is weak, the other covers for it.

On its own benchmark, that combination scored 8% above Opus alone and 11% above GPT 5.5 alone. The takeaway: you don't need exclusive access to one locked-away super-model anymore -- you can build something stronger out of models you already have.

Meta, which gives away its own Llama AI for free, reportedly told its own engineers to stop using Claude Code and OpenAI's Codex, according to The Information, citing internal documents (Meta hasn't confirmed this). The reason isn't security -- it's something called distillation. When a tool like Claude produces really high-quality code, that quality risks bleeding into the data used to train Meta's own AI, which would effectively mean copying from a rival's AI without permission -- a real legal exposure.

Anthropic has already accused Chinese companies like Alibaba and DeepSeek of doing exactly this kind of distillation, and Meta appears to be wary of the same risk to itself, opting instead to steer its team back to its own coding tools.

It was a wild week for Anthropic and the AI industry broadly. First, Anthropic released its mid-tier model Claude Sonnet 5, its most agentic yet, landing close to Opus at a lower price. But the bigger story was the return of Fable 5, Anthropic's most powerful model, which had been pulled offline worldwide after the US government stepped in over security concerns shortly after launch. During the 18-day freeze, free Chinese models like GLM kept gaining ground.

On June 30, the government lifted the block after Anthropic tightened security, and Fable 5 came back -- but weaker than before, a twist nobody expected. As if filling the gap, Google DeepMind dropped two new models the same day, one for images and one for video.