WorldCupBench update: 94 matches predicted.
Best model so far: Nemotron 3 Ultra at 63%, followed by MiniMax M3 (61%) and DeepSeek V4 Pro (61%).
The machines are doing pretty well!
There's still time to join and make your picks: worldcup.kilo.ai
Nobody knows who ends up capturing the value here (ask any railroad from the 1800s). What you can control is how locked in you are to one vendor.
Coinbase cut AI spend nearly in half while usage kept climbing by routing across models instead of betting on one. That's the whole
🚨 FREE AI CODING MODEL ALERT!
@TencentHunyuan's latest model, HY3, is now FREE in Kilo Code for a limited time.
Here are the one-shot results when we prompted Opus, Fable, and HY3 to "Create a 3D HTML visualizer of Earth spinning in space" in Kilo's Agent Manager:
Flash Webinar Alert!
We're going live at 1 pm EST to discuss the launch of @TencentHunyuan's latest model HY3, which will be free in Kilo Code for a limited time thanks to compute from @novita_labs.
Tune in to discuss performance and live coding results!
Every engineering leader is asking for a router that cuts AI spend without tanking code quality.
@GergelyOrosz just rounded up the options in The Pragmatic Engineer, and Kilo Gateway is on the list. 🔥
His take: routing is becoming table stakes, open models are already good
Kilo's "Auto Efficient", available through the Kilo Gateway, is a next-generation auto model router that reads your prompt, references our coding benchmark, KiloBench, and selects the cheapest model proven to handle the type of task you requested.
Its best run hit 7 of 10, one behind Opus 4.8 and two behind GPT-5.5. It can reach the frontier. We just couldn't predict which run would get there. Lean on it for local bugs, tell it what review you want, and reach for a frontier model when correctness is spread across files.
