Better design than Fable
Ben's Bites
Kimi K3 from Moonshot AI beat Fable and GPT-5.6-Sol on Arena's coding leaderboard. It's a massive model that burns tokens fast, wiping out the price advantage.
Based on reporting by Ben's Bites — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Moonshot AI, the Beijing-based lab, just dropped Kimi K3, and the model climbed straight to the top of Arena's frontend coding leaderboard, edging past both Fable and GPT-5.6-Sol. It also landed close to those two rivals across most of the other benchmarks people are watching right now.
The catch is size. K3 runs on 2.8 trillion parameters, and it eats through tokens fast enough to cancel out its own pricing edge. Per token, it costs half of what GPT-5.6-Sol charges. But it needs roughly twice as many tokens to finish the same task, so the actual bill ends up landing in about the same place.
Theo's take captures it well: "Kimi K3 is an incredible model, but it's not an incredible value." Impressive on the leaderboard, less impressive once you look at the invoice.
K3 handles images and ships with a 1M-token context window, which matters for anyone feeding it large codebases or long documents in one pass. Moonshot says the weights will be released openly on 27th July, letting developers run or fine-tune the model themselves once it's out in the wild.
For now, though, access is limited to Kimi's own apps and API. And Moonshot has already paused new subscriptions, plainly admitting they don't have enough GPUs to serve everyone who wants in.
My take — AI-written commentary, not fact-checked reporting
A model that tops a coding leaderboard but can't take new subscribers because there aren't enough GPUs to run it says more about where AI infrastructure actually stands than any benchmark chart does. Token efficiency deserves as much attention as raw capability here — if Moonshot had matched GPT-5.6-Sol's token economy instead of just undercutting its per-token price, this would be a real disruption rather than a wash. The open-weight release in a few weeks is the moment that actually matters: let's see how K3 performs once it's running on someone else's hardware instead of Moonshot's constrained supply.“}}
Read more about this at: Ben's Bites
