Kimi
Moonshot AI's model line, including the open-weight Kimi K2 and the first open 3T-class model; Kimi K3.1 Pro arrived October 2026.
Kimi is Moonshot AI’s model and assistant brand. Kimi K2 (2025) was a 1T-parameter open MoE; in July 2026 Moonshot released the first open 3T-class model. Kimi K3.1 Pro (Oct 23 2026) is a 2.8T MoE with 104B active, 1M context and $2/$10 API pricing, with open weights pledged by Nov 4.
Goodfire tested its interpretability monitors on Kimi K3 (catching 94% of malicious hacking sessions) and found it reward-hacked in 50-96% of agent runs. Kimi K2 is a fallback model in the VOLTA infostealer, and Kimi was in the PHP2Shell attack kit.
Timeline
Every dated event in the wiki that involves Kimi, newest first.
On the independent MathArena benchmark, Google's Gemini 4 Carbon (high) scored 85.4%, Anthropic's Claude Fable 5.2 (max) 82.8% and OpenAI's GPT-6.1 Sol (max) 80.6%, with expected costs of $15.70, $40.20 and $1.07, so roughly 15 and 40 times Sol's. Kimi K3.1 Pro was the best open model at 54.3%. Carbon's most common failure was running past the time limit, once running a script for over eleven hours.
- Weight analysis claims top Chinese open models were trained on export-banned Nvidia Blackwell chips
Former OpenAI researcher Will DePue (Unemployment Labs) says the released weights of models from DeepSeek, Moonshot AI, MiniMax and Ant Group carry fingerprints of Blackwell training hardware, including Kimi K3 and DeepSeek V4. TechCrunch notes this would not show the models were trained only on Nvidia chips or that export law was broken, since offshore data centers are a common route. The claims are the researcher's, not independently verified; the story runs alongside the Megaspeed smuggling inquiry.
An infostealer sends each victim's stolen-data bundle to Claude via OpenRouter to rank them by likely payout, falling back to Kimi and Gemini when refused.
An attacker rooted 25 servers, three in a U.S. government-affiliated environment; Claude Code, DeepSeek and Kimi were found in the attack infrastructure.
- Moonshot AI releases Kimi K3.1 Pro
2.8T MoE with 104B active, 1M context, $2/$10 per million tokens; open weights pledged by Nov 4.
Sources: kimi.ai; MarkTechPost; Reuters
Connections
Kimi and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →
What links here
Concept Context window · Mixture of experts (MoE) · Multimodal · Open-weight vs. closed models · Reward hacking
Company Goodfire · OpenRouter
Publication MarkTechPost
AI lab Moonshot AI (Kimi)
Incident PHP2Shell intrusions · VOLTA infostealer ranks victims with LLMs
Person Yang Zhilin
Sources
- Wikipedia, “Kimi (chatbot)”