Breaking

Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Model That Sometimes Answers 'I'm Claude'

July 17, 2026 at 22:53 EDT

Foundation Models
Open Source
Software Dev & Coding

Chinese startup Moonshot AI has unveiled Kimi K3, a 2.8-trillion-parameter flagship model that pushes a homegrown Chinese system to near-frontier performance—even as an unusual quirk, in which the model answers "I'm Claude" when asked its name, draws attention. The model was released around July 16, 2026, and is available through the company's website, app and API. Fortune reported the launch as a milestone for Chinese AI reaching what it called Fable-level territory.

July 16, 2026 · Moonshot AI

Kimi K3 pushes Chinese AI to the frontier

A 2.8-trillion-parameter, open-weight flagship reaches near-frontier performance in coding and long-document work — and undercuts U.S. rivals on price.

2.8T

parameters (Mixture-of-Experts)

token context window

~57.1

Intelligence Index (~4th overall)

Frontend Code Arena score

Kimi K3 edges ahead of Claude Fable 5

1,679

Kimi K3

1,631

Claude Fable 5

INPUT

$3 / M tokens

OUTPUT

$15 / M tokens

Undercuts comparable Claude Fable 5 and GPT-5.6 offerings.

The "I'm Claude" quirk

Asked its name in some tools, Kimi K3 may reply "I'm Claude, a helpful assistant."

Not genuine impersonation — an artifact of the system prompt used by integrations like Claude Code. Developers treat it as an amusing footnote, not a red flag.

STRENGTHS

Open weights (full release by Jul 27, 2026)
Strong frontend, 3D and long-form work
Some developers switching from Claude

LIMITS

Local run needs dozens of high-end GPUs
Practical deployment leans on the cloud
Durable everyday edge still unproven

Continue reading

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