Grok 4.7
Step 5 Preview
2.8T
- Context
- 131Ktokens
- Released
- 1y ago
- Size
- 1Tparameters
Kimi K2 Instruct is Moonshot AI's post-trained version of its Kimi-K2-Base foundation model, which the company says scales up its earlier Moonlight model. The launch announcement, titled 'Open Agentic Intelligence,' open-sources a model the maker says is optimized for agentic tasks, acting on them rather than only responding. It is built for general-purpose chat and agentic work, including tool use and coding.
How it compares
Level
FlagshipThe lab's most capable generally available line, priced for hard tasks.
Flagship models from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inAdditional source evidence
- Source model
- moonshotai/kimi-k2-instruct
- Source family
- kimi-k2
- Last updated
- 2025-07-11
- Model context
- 131K
- Model output limit
- 33K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- No
- Tool calling
- Yes
- Temperature control
- Yes
Downloadable weights
Complete source record (JSON)
{
"id": "moonshotai/kimi-k2-instruct",
"name": "Kimi K2 Instruct",
"description": "Non-thinking Kimi K2 MoE instruction model for coding and tool-using agents",
"family": "kimi-k2",
"last_updated": "2025-07-11",
"attachment": false,
"reasoning": false,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 131072,
"output": 32768
},
"release_date": "2025-07-11",
"open_weights": true,
"license": "Modified MIT",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct"
}
]
}Frequently asked questions
Running locally
Kimi K2 Instruct needs an estimated 716.1 GB of VRAM at Q4, 1,254.6 GB at Q8 and 2,361.2 GB at FP16, based on its 1T parameters. These are estimates: weights plus about 15% and 0.5 GB for a short context.
No. A 24 GB, 16 GB, 12 GB or 8 GB GPU does not fit Kimi K2 Instruct at Q4, Q8 or FP16 (estimate).
Privacy and data
Yes. Kimi K2 Instruct has open weights, so you can self-host it for full control of your data.
Capabilities and specs
Kimi K2 Instruct has a context window of 131K tokens (131,072) and a maximum output of 33K tokens (32,768).
No. Kimi K2 Instruct accepts text only and cannot read images, audio, video or PDFs.
Yes. Kimi K2 Instruct supports tool calling (function calling).
No. Kimi K2 Instruct does not have a reasoning or thinking mode.
Quality and benchmarks
FlexOlmo 7x7B by Ai2, Devstral Small by Mistral AI and DiffuCoder 7B by Apple. They are matched on model kind and release date, with no comparable price.
FlexOlmo 7x7BKimi K2 Instruct vs FlexOlmo 7x7BDevstral SmallKimi K2 Instruct vs Devstral SmallDiffuCoder 7BKimi K2 Instruct vs DiffuCoder 7B
Versions and lineage
Both have a 131K token (131,072) context and both were released on July 11, 2025. The difference is that Kimi K2 0711 has a price of $0.57 input and $2.30 output per 1M tokens.
Licensing and access
Kimi K2 Instruct is open weights, released under the Modified MIT license.
The Kimi K2 Instruct weights are available from the Hugging Face repository moonshotai/Kimi-K2-Instruct.
Basics
Kimi K2 Instruct was made by Moonshot AI and released on July 11, 2025.
Yes. Kimi K2 Instruct is open weights.
Generated from Vertical Tabs data, updated November 10, 2026.
Discussion
No comments yet
Be the first to share your thoughts.