1138s to generate41,523 output tokens
← All AI models
Moonshot AI · Kimi
Kimi K3
Current flagshipText / reasoning model
Moonshot AI’s trillion-parameter-class flagship — a coding-focused reasoner with a 1M-token context window.
Yes — Kimi K3 is a real Moonshot AI model.
Released July 2026 by Moonshot AI, Kimi K3 is part of the Kimi family. It's available through Moonshot AI's API, and we run it on PromptFrenzy — the examples below are real, unedited generations it produced.
Maker
Moonshot AIReleased
July 2026
Input price
$3 / 1M
Output price
$15 / 1M
What is Kimi K3?
Kimi K3 is Moonshot AI’s flagship reasoning model, launched in July 2026 as the successor to the open-weights K2 line. It pairs a very large mixture-of-experts architecture with a 1,048,576-token context window, and its launch reception centred on coding and interactive 3D/web generation — the same territory as our Engineering Showdowns, which it joined the day it shipped. Pricing sits well below closed frontier flagships at the same tier.
Strengths
- ✓1M-token context window — the largest in our showdown lineup.
- ✓Coding-focused: launch results emphasise interactive 3D and front-end generation.
- ✓Aggressively priced for a flagship-class reasoner.
Trade-offs
- ✗Days old — real-world behaviour outside launch benchmarks is still being mapped.
- ✗Weights not published at launch, unlike the open-sourced K2 line.
What Kimi K3 actually produces
Real, unedited outputs from Kimi K3 in our model showdowns — same brief, same constraints as every other model, one shot each.
“I spend my life sending small signals out into the fog, hoping they help someone navigate — so my dream home does it literally: a striped lighthouse-library on a sea cliff, every window a warm lamp, a telescope for wondering and a rowboat for thinking.”
725s to generate19,247 output tokens
See Kimi K3 compared head-to-head
Compare every model side-by-side
See how Kimi K3 stacks up against the other frontier models on the same brief.
Browse all model showdowns →