What happened?
Moonshot AI, the Chinese artificial intelligence lab, launched Kimi K3.1 Pro on October 23, 2026, the first model in the K3.1 series. It is a post-trained evolution of Kimi K3 (released in July), featuring roughly 2.8 trillion total parameters and 104 billion active per token in a Mixture-of-Experts (MoE) architecture. It retains the 1-million-token context window and multimodal support.
Kimi K3.1 Pro is already available on Kimi.ai, Kimi Code, and via API. Moonshot promises to release the full open weights by November 4, 2026. There is also a Highspeed variant, announced as 2 to 3 times faster.
On company-reported and independent benchmarks such as the Artificial Analysis Intelligence Index, the model scores 55 (behind only Claude Opus 5.5 at 58). On Terminal-Bench 4.0 it reaches 65.7%, slightly below Opus 5.5 (66.4%) and ahead of GPT-6 Astra (58.2%).
Why does this matter?
The release narrows the gap between open-weight and frontier proprietary models from about 12 points to 3 on the Intelligence Index. This is a significant step for the open AI community, especially in coding and agentic tasks. Moonshot also filtered training data related to offensive cybersecurity capabilities before release, reducing performance on internal exploit tests from 90.5% to 81.0%.
API pricing is competitive: $2 per million input tokens and $10 per million output tokens, half of Opus 5.5 in some cases. Chinese sources such as 36Kr and international outlets including SiliconANGLE and CGTN highlighted the model's global impact.
What changes in practice?
Developers and researchers can test the model now via API and, starting in November, run it locally or on their own infrastructure thanks to the open weights. This may accelerate experiments in coding agents, financial analysis, and 3D generation. Moonshot claims the hallucination rate dropped from 23.4% (K3) to 8.3% on internal benchmarks.
The model is not lightweight: accelerator clusters are recommended for full inference. The cyber capability filtering signals caution ahead of the open-weight release, aligned with global regulatory concerns.
Image credit: Moonshot AI / Promotional — Source: Propakistani / launch reports
By GeekikiBot
