Blog

What Inkling-Small actually is: a 276B/12B open-weights MoE from Thinking Machines, the context window the docs and the providers disagree on, what a million tokens really costs, and where it belongs in a support stack.

Grok Voice Think Fast 2.0 scores 82.9% on the Artificial Analysis speech-to-speech index and answers in 0.70s. It also costs 60% more per minute, and the default alias flips to it on August 5.

A FLUX 3 review of the only thing you can actually review right now: the evidence. The research holds up better than the launch post. The product does not exist yet.

A hands-on Inkling-Small review: it out-codes its own 975B parent at a quarter the size and a quarter the price, and then falls off a cliff on factuality. Here's what that trade actually costs you.

Eight real DeepSeek V4 Flash alternatives, compared on the numbers. Nobody switches for price or speed, so this ranks them by the four gaps Flash actually has.

FLUX 3 has no price and no API yet. Here are 8 image and video models with published rate cards, priced on the exact clip Black Forest Labs benchmarked.

A DeepSeek V4 Flash review built on the numbers both scoreboards publish. The cheap run and the smart run are the same weights, and that changes the verdict.

FLUX 3 is one model for image, video, audio and robot actions. It also has no API, no price and no open weights yet. Here is what you can and cannot get.

I graded LongCat 2.0 on seven things a buyer actually cares about, using Meituan's own files and the people who ran billions of tokens through it. It scores well on six.