✳ the wire · intel
Kimi K3 inference: Fireworks Fast is 4.5× the throughput at 1.5× the price, and uptime is the real spread
An email when Moonshot’s next model answers on a public API, about 15 minutes after members, plus every frontier launch · no card
The provider table for Kimi K3 is unusually wide right now, and the cheapest row is not the interesting one.
Most providers — Modal, Baseten, Together, Moonshot itself, DigitalOcean, Fireworks standard — sit at the same $3.00 / $15.00 per 1M with $0.30 cache reads. Within that identical price, P50 latency runs from 3.32s (Modal) to 7.29s (Moonshot's own endpoint), and throughput from 17 tps (Baseten) to 36 tps (Modal). Same model, same price, better than 2× the speed depending on who you route to.
Every wire item is source-verified before it posts and graded: confirmed · strong · reported · rumor. Subscribers see the primary source on each one, the reasoning behind its grade, and the full analysis — plus the launch alerts themselves, which fire the minute a model answers rather than whenever the feed catches up.
