Kimi K2.6
Moonshot AI's March 2026 frontier update. Continued gains in agentic benchmarks and tool use. Among the most-pulled models in 2026.
What Kimi K2.6 is for
Kimi K2.6 is a large mixture-of-experts model: only a fraction of its parameters activate per token, but all of them have to be resident, so total size decides whether you can run it at all. Moonshot AI published it in 2026-03 under Modified MIT with no published dense parameter count and 128K of context. The catalogue records its strength as Kimi K2.6 — March 2026 update.
Sizes and memory
Our catalogue does not carry a per-size memory breakdown for Kimi K2.6, because no dense parameter count has been published for it. What it records instead is a floor of 256 GB and a recommended 512 GB for the default configuration. Until a per-size figure exists, treat those two numbers as the whole guidance rather than reading a size table that would have to be invented.
Hardware that runs Kimi K2.6
Without a per-size memory figure there is no honest way to say which card holds Kimi K2.6. Its recorded floor of 256 GB rules out every consumer card in our hardware guide, which tops out well below that.
What people use it for
The catalogue lists 3 recorded uses for Kimi K2.6: agentic, long-context, tool use. It is tagged MoE, tools, agentic, long-context.
Where Kimi K2.6 sits in the Kimi family
Our catalogue holds 4 Kimi entries. Ordered by release date, Kimi K2.6 (2026-03) is the 3rd of them, and Kimi K2.7 Code (2026-06) is newer. Kimi K2.5 and Kimi K2.7 Code carry the same recorded memory footprint, so choosing between those and this one is a question of behaviour rather than of hardware: Kimi K2.6 is the one recorded here for Kimi K2.6 — March 2026 update.
It ships under the Modified MIT, which is neither a standard open-source licence nor a closed one. Read the terms before shipping anything commercial on top of it.
Context window: 128K
128K tokens is roughly 96,000 words at the usual ratio, which is a small repository or a whole book. Two cautions come with a window this size: the KV cache at full length can rival the weights for memory, and models rarely use the far end of their advertised context as well as they use the near end. Treat it as headroom, not as a promise.
Read the Kimi K2.7 Code page instead
Our catalogue records the same publisher, the same mixture-of-experts configuration, the same 256 GB memory floor and the same 128K context window for Kimi K2.5, Kimi K2.6, Kimi K2.7 Code. Where we cannot show you how two releases differ in anything a reader would act on, publishing both as competing search results would be a guess dressed as a recommendation.
So Kimi K2.7 Code (2026-06), the most recent of the 3, is the one we put in front of that question, and this page is marked noindex. It stays published because the record is real and someone running Kimi K2.6 today should still be able to look it up. When a figure appears that separates them, this page returns to the index on its own.
How to run Kimi K2.6 locally
Our catalogue does not record registry tags, so look the current tag up in the Ollama library or on Hugging Face before pasting these. The commands below are the shape of the workflow, not a copy-and-paste recipe.
Option 1: Ollama (simplest)
# find the tag for Kimi K2.6 at ollama.com/library
ollama run <tag>Option 2: Mullama (production)
mullama pull <tag>
mullama run <tag>Option 3: llama.cpp (CLI)
# download a GGUF from Hugging Face, searching for "Kimi K2.6 GGUF"
./llama-cli -m model.Q4_K_M.gguf -p "Hello, AI!"Option 4: Python with Mullama or llama-cpp-python
from mullama import Model, Context
model = Model.load("model.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256))Sources
Model record from src/data/models.json (2026-06-29); card capacities from src/data/gpus.json (2026-06-29). Upstream: Curated from ollama.com/library + community benchmarks + paperswithcode + huggingface.