Breaking

Moonshot AI Unveils Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model, Reigniting the US-China AI Race

July 20, 2026 at 07:56 EDT

Foundation Models
Open Source
Software Dev & Coding

Chinese startup Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight model it bills as the world's first 3-trillion-parameter-class open-source system, intensifying debate over how quickly China is closing the gap with US AI leaders. The model is available immediately through Moonshot's API, with full model weights scheduled for release on July 27, 2026.

July 17, 2026 · Moonshot AI

Kimi K3: China's First 3-Trillion-Parameter Open-Weight Model

A 2.8T-parameter Mixture-of-Experts system with a 1M-token context, shipped at aggressive prices despite US chip controls — a strong entry, though not yet a runaway leader.

2.8T

Parameters (MoE: 16 of 896 experts active per token)

Token context window for long coding & reasoning

93.5%

GPQA Diamond (graduate-level science reasoning)

Price per 1M tokens vs. the frontier field

Observers estimate Kimi K3 runs ~40–50% cheaper than comparable frontier models.

$0.30

Input
(cache hit)

$3.00

Input
(cache miss)

$15.00

Output

Where it lands on independent benchmarks

Frontend Code Arena

Edges out Claude Fable 5 on iterative front-end coding

Intelligence Index — score 57

On par with Opus 4.8 & GPT-5.5; behind Fable 5 and GPT-5.6 Sol

The upside

A top-tier open-weight model at aggressive prices, strong on long coding tasks — front-end, game dev and kernel optimization — shipped despite chip constraints.

The caution

Overall performance still trails Fable 5 and GPT-5.6 Sol, with user-experience gaps remaining — a strong entry, not a proven turning point.

Full model weights are scheduled for release on July 27, 2026 — letting independent teams download, fine-tune and deploy Kimi K3 at scale, and testing whether it marks a turning point or simply a strong step in a fast-moving pack.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.