In which areas is the Kimi K3 model more successful than Claude Fable 5?

According to the data shared by the company, Kimi K3 scored higher than Claude Fable 5 in coding, terminal operations, web research, automation, and spreadsheet tasks.

What are the current usage limitations of the Kimi K3 model?

The model can make unexpected decisions by falling behind its competitors in some tests, such as visual reasoning. Additionally, since the full source weights of the model have not yet been released, it is closed to independent review.

When was Kimi K3 released and how is it accessible?

The model was released on 16 July 2026 and is accessible via Kimi's website, mobile apps, coding tool, and API.

一覧に戻る

China's Moonshot AI introduces 2.8 trillion-parameter Kimi K3 model

News9 Live

要点

Moonshot AI launched the Kimi K3 model, featuring 2.8 trillion parameters and a 1 million context window.
In company tests, the model outperformed Claude Fable 5 in areas such as coding and web research.
Kimi K3 can autonomously execute long engineering tasks for 15 hours with limited human oversight.
The full weights of the model will be released on 27 July 2026 and opened for independent testing.

数字で見る

2.8 trillion parameters1 million context window15 hours of autonomous process27 July release date

Chinese artificial intelligence company Moonshot AI announced its new Kimi K3 model with 2.8 trillion parameters. Designed for long-duration tasks such as coding, research, and office automation, the model was made available on 16 July. In the benchmark tests announced by the company, Kimi K3 outperformed Anthropic's Claude Fable 5 model in several critical areas.

The new model was observed to score higher than its competitors in coding and web research tests such as Terminal Bench, SWE Marathon, and BrowseComp. Additionally, featuring a context window of 1 million, Kimi K3 can manage very large documents and long-term engineering tasks in a single session. The company stated that the model can sustain operations for up to 15 hours without human intervention.

Despite this, Moonshot AI acknowledges that the model's overall performance still lags behind Claude Fable 5 and OpenAI's models. Kimi K3 has limitations in some areas, such as visual reasoning, and can make unexpected decisions in ambiguous tasks. The full weights of the model are planned to be released on 27 July 2026.

リアクション

#Moonshot Ai#Kimi K3#Large Language Models#Artificial Intelligence#China#Anthropic

この記事について質問

回答はこの記事のみからAIが生成します。

よくある質問

In which areas is the Kimi K3 model more successful than Claude Fable 5?: According to the data shared by the company, Kimi K3 scored higher than Claude Fable 5 in coding, terminal operations, web research, automation, and spreadsheet tasks.
What are the current usage limitations of the Kimi K3 model?: The model can make unexpected decisions by falling behind its competitors in some tests, such as visual reasoning. Additionally, since the full source weights of the model have not yet been released, it is closed to independent review.
When was Kimi K3 released and how is it accessible?: The model was released on 16 July 2026 and is accessible via Kimi's website, mobile apps, coding tool, and API.