by Moonshot AI
Moonshot AI's open-weight 2.8T-parameter multimodal MoE agentic model with 1M-token context.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Kimi K3.1 Pro is a Mixture-of-Experts model from Moonshot AI with 2.8T total and 104B activated parameters, 93 layers (69 KDA and 24 Gated DSA attention layers) and a 1-million-token context window. It is natively multimodal (image-text-to-text) and post-trained from Kimi K3 with large-scale reinforcement learning. It is the first model in the K3.1 series.
Moonshot positions it for coding, 3D creation, financial research, factual accuracy and other agentic tasks. Creators report benchmark results near top closed models (e.g. Terminal Bench 65.7, OSWorld 79.6), though one hands-on review found real-world tests fell short of the best closed models.
It is available through the Kimi chat site and the Kimi platform API (reported at about $2 input / $10 output per million tokens). Weights are published on Hugging Face under a custom "kimi-k3.1-pro" license in compressed-tensors format. The checkpoint is roughly 1.5 TB, so self-hosting needs on the order of 1.2 TB of VRAM, which limits it to enterprise-scale infrastructure.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Best-performing open model; weights now released on Hugging Face. 2.8T total parameter MoE with 104B active, 1.56 TB in size, aimed at enterprises.
Top open-weight model, just below Claude Fable 5.1, with weights now on Hugging Face: 2.8 trillion parameters, 1.56 TB.
Moonshot released open weights for Kimi K3.1 Pro, a roughly 3T-parameter frontier-grade model needing about 1.2 TB VRAM. It ranks near the top on Terminal Bench, Agent SL Exam, OS World and GDP Eval, just behind Opus 5.5 in several.
Open-weight model with 2.8T total parameters (104B active) and 1M context, weights due November 4th. It scores 46 on Artificial Analysis, well above other open models, and ranked 11th on his bench. It's impractical to run at home but matters for on-premise use.
Newest Moonshot model, scores higher overall on benchmarks, but the creator thinks DeepSeek is better. Its iPhone Duo 3D test was worse than DeepSeek's.
3 trillion parameter model with a 1M context window, roughly on par with frontier models on benchmarks and cheaper per task. Available on Kimi's platform and API, with open weights planned for release later.
Open-weights frontier-class MoE model (~2.8T total, 104B active, 1M context, multimodal). The creator tested it on a Rubik's Cube sim (strong pass), iPhone 3D render, Blender knot (fail), motion graphics, a Fall Guys clone and a site redesign. Benchmarks are near Opus 5.5 (Terminal Bench 65.7, OSWorld 79.6, AA index 55), priced at $2/$10 per million tokens, but real-world tests fall short of the top closed models.
Closest neighbours by meaning, not just by category.
Model · Moonshot AI
Moonshot AI's open-weights 1T-parameter MoE multimodal agentic model with 256K context, vision input and agent swarm.
Model · Moonshot AI
Moonshot AI's Kimi K3 model family, with K3.1 Pro as the current release in the Kimi app and API.
Model · Moonshot AI
Moonshot AI's large language model, described as open-source with roughly 2T+ parameters, benchmarked by Artificial Analysis.
Model · Moonshot AI
Moonshot AI's Kimi K2.7 coding model, offered through the Kimi Code agent (CLI, VS Code extension, desktop app).
Model · MiniMax
Open-weights native multimodal MoE model from MiniMax, ~428B parameters (~23B active) with 1M-token context.
Model · Meituan
Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.