The artificial intelligence race between Moonshot's new K3 model and Claude Code
要点
- 1. Moonshot AI announced the Kimi K3 model, featuring 2.8 trillion parameters and a 1 million token context window, on 16 July 2026.
- 2. With a pricing of 3/15 dollars per millionth token, K3 has the exact same cost as Claude Sonnet 4.6.
- 3. In the independent test Artificial Analysis Intelligence Index, K3 scored 57 points, showing similar performance to Claude Opus 4.8.
- 4. Alibaba's ban on Claude Code due to the detection of a hidden code triggered trust issues regarding artificial intelligence tools.
数据一览
Moonshot AI's 2.8 trillion parameter Kimi K3 model, released on 16 July 2026, has shifted the balance in the coding assistant market. The open-weight model with a 1 million token context window scored 57 points on the Artificial Analysis Intelligence Index, demonstrating a performance on par with Claude Opus 4.8.
The new model is positioned in the exact same price tier as Claude Sonnet 4.6, with a pricing of 3 dollars per input and 15 dollars per output. This situation eliminated the massive cost advantage of approximately 84 percent that the previous Kimi K2.7 Code version offered compared to Claude Opus 4.8, bringing it to the Sonnet level. Developers are now evaluating factors such as open-weight advantages, data residency, and ecosystem integration instead of cost when choosing between these two models.
Geopolitical and security-focused developments in the background of the competition also draw attention. Alibaba banned Claude Code from its systems starting from 10 July due to the detection of a hidden monitoring code. Additionally, Anthropic's most powerful model, Claude Fable 5, could only be made available again globally on 1 July following a 19-day restriction due to US export controls.
对这条新闻做出反应
询问这条新闻
回答由AI仅根据本新闻生成。
常见问题
- 1. S: What is the price difference between Kimi K3 and Claude Sonnet 4.6?
- C: There is no price difference; both models are priced exactly the same, at 3 dollars for input and 15 dollars for output per millionth token.
- 2. S: Is Kimi K3 more successful than Claude Opus 4.8 in independent tests?
- C: According to Artificial Analysis data, K3 has a performance comparable to Opus 4.8's (56) with a score of 57 points. However, some of K3's superior benchmark results come solely from Moonshot's own tests.
- 3. S: Why did Alibaba ban Claude Code?
- C: After researchers found a hidden monitoring code within the code that identifies Chinese users, Alibaba banned Claude Code from its internal systems effective from 10 July.