Moonshot AI Launches Kimi K2.6, a 1T-Parameter Open-Weight Model

aiai-modelingai-products 2 posts · 2 accounts

Moonshot AI has released Kimi K2.6, a 1 trillion-parameter open-weight model designed to coordinate 300 parallel agents across 4,000 steps. The model is available through the company’s platform and API.

The announcement says Kimi K2.6 scored 54.0% on Humanity’s Last Exam, ahead of GPT-5.4 and Claude Opus 4.6 on that benchmark.

From the sources (2 posts)

@nandodf

RT @ArtificialAnlys: Moonshot’s Kimi K2.6 is the new leading open weights model. Kimi K2.6 lands at #4 on the Artificial Analysis Intellige…

@aiwire_x

Moonshot AI just dropped Kimi K2.6 — a 1T-parameter open-weight model that coordinates 300 parallel agents and 4,000 steps. It scores 54.0% on Humanity's Last Exam, outpacing GPT-5.4 and Claude Opus 4.6. Available now on

Market impact

72 big

NVDA MSFT GOOGL SOXX QQQ AMZN

Why it matters: A new leading open-weight frontier model that beats GPT-5.4 and Claude Opus 4.6 on a top benchmark threatens the closed-lab margin thesis and the AI-capex return case, repricing the largest index weights (NVDA, closed-lab proxies) as a DeepSeek-style event.

At publish time: Headline states a 1T-parameter open-weight model beating GPT-5.4 and Claude Opus 4.6 on Humanity's Last Exam and ranking #4 on Artificial Analysis — an open-weight frontier capability claim, tradable at publish time.

model confidence: medium

What happened next

1–2 weeks later

fizzled moved up

No DeepSeek-style repricing; NVDA/megacaps kept rising, hitting all-time high May 14 as hyperscaler capex accelerated. K2.6 follow-ups were benchmark chatter, not a market event.

based on follow-up coverage

Follow-up coverage

No follow-up coverage matched this story.

DeepSeek V4 Pro Climbs to No. 2 Among Open-Weight Models US Sanctions China's Hengli and Tightens Iran Oil Curbs