Understand this story
On November 4, 2026, Moonshot AI released the model weights of Kimi K3.1 Pro. The next day, Kimi.ai officially announced the release of the model weights along with a link to obtain them, and stated that it will continue to advance open intelligence. According to Arena statistics, Kimi K3.1 Pro ranks 1st among open-source models in Agent Arena and 13th on the overall leaderboard; these statistics are based on 17.5K+ real long-horizon agent sessions. Its Max version shows a net improvement of +5.1%, ranking first among open-source models in all three categories of Work, Code, and Chat, an increase of 4.2 percentage points over Kimi K3 (Max). By sub-item, it ranks first among open-source models in three items—Bash Reliability (+7.9%), Praise vs Complaint (+7.7%), and Steerability (+6.5%)—and ranks 4th among open-source models in Confirmed Success (+3.9%).
AI-generated from public reports
Latest development · Report
Moonshot AI Open-Sources Kimi K3.1 Pro Weights, Tops Agent Arena Open-Source Model LeaderboardDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 5
ArenaSelected
Moonshot AI Open-Sources Kimi K3.1 Pro Weights, Tops Agent Arena Open-Source Model Leaderboard
Moonshot AI released the model weights of Kimi K3.1 Pro, which ranks 1st among open-source models and 13th overall on Agent Arena. Based on statistics from 17.5K+ real long-horizon agent sessions, Kimi K3.1 Pro (Max) shows a net improvement of +5.1%, ranking first among open-source models in all three categories of Work, Code, and Chat, and improving 4.2 percentage points over Kimi K3 (Max). In sub-scores, it ranks first among open-source models in Bash Reliability (+7.9%), Praise vs Complaint (+7.7%), and Steerability (+6.5%), and 4th among open-source models in Confirmed Success (+3.9%).
Testing CatalogSelected
Kimi K3.1 Pro weights are now open on Hugging Face under the eponymous license
The model weights of Kimi K3.1 Pro have been opened on Hugging Face under the "Kimi K3.1 Pro" license. According to Kimi's official statement, the model is an open-weight, natively multimodal agent model, and also its strongest 3T-scale model to date.
Rohan PaulSelected
Moonshot AI releases 2.8T-parameter Kimi K3.1 Pro weights on Hugging Face
Moonshot AI released the model weights of the 2.8T-parameter Kimi K3.1 Pro on Hugging Face, trained and released in 4-bit precision to avoid the 5.6TB storage required for a 16-bit version. In official tests, the model scored 65.7% on Terminal-Bench 4.0, close to Claude Opus 5.5's 66.4%, but without tools it scored only 39.4% on HLE-Diamond, below GPT-6 Astra's 65.8%. The model activates only 104B parameters per token, and 69 of its 93 layers use linear attention to reduce the processing cost of a 1M-token context; a modified MIT license permits internal use, but model-hosting companies with annual revenue over $20 million must negotiate separately.
CuiMaoRelated discussion
CuiMao joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Kimi.aiFirst-partySelected
Kimi Releases K3.1 Pro Model Weights
Kimi announced the release of the model weights for Kimi K3.1 Pro, along with a link to obtain them, and stated that it will continue to advance open intelligence.
Nov 4
Hacker News TrendingSelected
Moonshot AI Releases Kimi K3.1 Pro Open-Source Multimodal Agent Model
Moonshot AI has released Kimi K3.1 Pro, the first model in the K3.1 series. It uses a MoE architecture with 896 experts, activating 16 per token, supports a 1-million-token context window, and its weights are open-sourced in MXFP4 quantization. The model was post-trained from Kimi K3 via large-scale joint reinforcement learning across hundreds of downstream domains, and shows clear improvements in agentic capabilities such as coding, 3D creation, financial research, and factual accuracy. The API is now live at platform.kimi.ai under the model name kimi-k3.1-pro, offering OpenAI/Anthropic-compatible interfaces, with reasoning_effort supporting three levels—low, high, and max—defaulting to max.
X.PINSelected
Kimi K3.1 Pro proactively weakens attack capabilities; China plans to include large models in export controls
When releasing Kimi K3.1 Pro, Moonshot AI removed cyberattack-related content from its training data and reinforcement learning tasks; in internal tests, the pass rate for exploiting software vulnerabilities dropped from 90.5% to 81.0%, in order to reduce the risk of misuse after open-weight release. On October 16, China's Ministry of Commerce released a draft update to the catalog of technologies subject to export restrictions, planning to include large models, requiring companies to register in advance, with open-source projects eligible for expedited approval. The article also mentions that when relevant UK and US institutions previously tested Kimi K3, they found that its safety rules failed to block attack tasks, but its capabilities still lagged significantly behind leading US models.
Heat trend for this story
- Current
- 53
- Period peak
- 58 Nov 5 06:00
- 24-hour change
- —
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.