Fable Is Back — But With Caveats

The most consequential AI news of the week is that Anthropic's Fable 5 is coming back online after 19 days under government export controls. But the return comes with new guardrails, unanswered questions about what exactly changed behind the scenes, and a narrower window of free access than users might have hoped. Meanwhile, OpenAI appears to have found a way to cut inference costs in half for its least engaged users, Anthropic quietly launched a new Sonnet model that has the community divided, and AWS is spending a billion dollars to build a forward-deployed engineering division.

The Inference Cost Race Heats Up

OpenAI has reportedly discovered an optimization technique that cut inference requirements in half for existing models — at least for a specific subset of users. When applied to ChatGPT users who weren't signed into the service, OpenAI was able to serve that entire segment on just 100 GPUs. The technique itself has not been disclosed, even to other OpenAI employees. The Information's Stephanie Palazzolo described it as "a very important secret sauce for them that they don't even want to tell other OpenAI employees about, because if these things leak, it can quickly be picked up by other labs."