Kimi K3.1 Pro Released: Moonshot AI's Strongest 3-Trillion-Parameter Model with 1M Context (Oct 24, 2026)

[ikaijua]

On October 24, 2026, Moonshot AI officially released Kimi K3.1 Pro — the first model in the K3.1 series and the company's strongest model to date: a 3-trillion-parameter model with a 1M-token context window. A high-speed variant, Kimi K3.1 Pro Highspeed, ships with the same intelligence at roughly 2-3x output speed.

Where to use it

• kimi.com / Kimi App / Kimi Work / Kimi Code
• API: platform.kimi.com
• Full model weights to be released by November 4, 2026

Highlights

Area	Result
Coding agents	Terminal-Bench 4.0: 65.7% (GPT-6 Astra 58.2%, Claude Opus 5.5 66.4%); internal Kimi Code Bench 3.0 Long-horizon: 61.2% (up from K3's 55.95%)
3D creation	BenchCAD: 92.7% — writes and runs Blender/CAD scripts directly for parametric modeling; new 3D Make feature on kimi.com and the app (preview, edit, export for 3D printing / video / live photos)
Finance workflows	OfficeQA Pro v2: 64.4 — current SOTA; internal finance benchmarks: financial modeling +14%, financial research +20%, business operations ~4.2x K3
Factuality	Kimi Hallu Bench Hard statement-level error rate cut from 23.4% (K3) to 8.3%
Game dev / frontend	Fluent with Unity, Unreal Engine, and Godot; noticeably better Three.js scenes and 3D assets

Feel free to share hands-on experiences, benchmark comparisons, and related tooling worth adding to Awesome-AITools.

Chinese version: Kimi K3.1 Pro 正式发布

---

Sources: Kimi API Platform — K3.1 Pro Quickstart, IT之家 / ifeng Tech report