Kimi K3

Reading a model's internals instead of its output cut the cost of watching for cyber misuse by about thirty times, Goodfire says

A lightweight probe reads activations at every turn and escalates only what looks wrong, which drops judge cost from $3,166 to $65 per million turns.

A German lab put 967 sensitive prompts to six Chinese open models, and one answered 17% of them evenly

An American model was flagged too. Aleph Alpha traced it to about 3,500 rows of a 9.3-million-row fine-tuning set that carried Communist Party framing.

Harvey raises $550m at a $15.5bn valuation, six months after it was worth $11bn

The raise follows Tenet, a model Harvey post-trained from Moonshot's open-weight Kimi K3, and nearly doubles the company's value in nine months.

A coding benchmark built on licensed private codebases puts Fable 5.1 first at 38.8%

The scores come from the benchmark maker's own runs, not from the model vendors, and so far they rest on ten tasks, each attempted eight times per model.

Moonshot AI tells investors it can reach $2bn in annualised revenue by the end of the year

The growth is real and the margins are not the same as a closed lab's. Moonshot gives its model weights away, which is what got it to this revenue and what caps the profit on it.