What are the hardware requirements for K2 Merged 3.5T Fp8?

K2 Merged 3.5T Fp8 requires approximately 392.2 GB of VRAM and supports a context window of 128K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed K2 Merged 3.5T Fp8 and how large is it?

K2 Merged 3.5T Fp8 is developed by NousResearch. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate K2 Merged 3.5T Fp8?

K2 Merged 3.5T Fp8 is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

K2 Merged 3.5T Fp8 by NousResearch — VRAM 392.2GB, 128K context

Name: K2 Merged 3.5T Fp8
Author: NousResearch

K2 Merged 3.5T Fp8 is an open-source language model by NousResearch. Features: LLM, VRAM: 392.2GB, Context: 128K, License: other, Merged, LLM Explorer Score: 0.21.

Merged Model Conversational Custom code Deepseek v3 Endpoints compatible Fp8 Region:us Safetensors Sharded Tensorflow

Model Card on HF 🤗: https://huggingface.co/NousResearch/k2-merged-3.5T-fp8

K2 Merged 3.5T Fp8 Parameters and Internals

LLM Name	K2 Merged 3.5T Fp8
Repository 🤗	https://huggingface.co/NousResearch/k2-merged-3.5T-fp8
Merged Model	Yes
Required VRAM	392.2 GB
Updated	2026-08-01
Maintainer	NousResearch
Model Type	deepseek_v3
Model Files	12.3 GB: 1-of-348 10.0 GB: 2-of-348 10.0 GB: 3-of-348 10.0 GB: 4-of-348 10.0 GB: 5-of-348 10.0 GB: 6-of-348 10.0 GB: 7-of-348 10.0 GB: 8-of-348 10.0 GB: 9-of-348 9.9 GB: 10-of-348 10.0 GB: 11-of-348 10.0 GB: 12-of-348 10.0 GB: 13-of-348 10.0 GB: 14-of-348 10.0 GB: 15-of-348 10.0 GB: 16-of-348 10.0 GB: 17-of-348 10.0 GB: 18-of-348 10.0 GB: 19-of-348 10.0 GB: 20-of-348 10.0 GB: 21-of-348 10.0 GB: 22-of-348 10.0 GB: 23-of-348 10.0 GB: 24-of-348 10.0 GB: 25-of-348 10.0 GB: 26-of-348 10.0 GB: 27-of-348 10.0 GB: 28-of-348 10.0 GB: 29-of-348 10.0 GB: 30-of-348 10.0 GB: 31-of-348 10.0 GB: 32-of-348 10.0 GB: 33-of-348 10.0 GB: 34-of-348 10.0 GB: 35-of-348 10.0 GB: 36-of-348 10.0 GB: 37-of-348 10.0 GB: 38-of-348 10.0 GB: 39-of-348
Model Architecture	DeepseekV3ForCausalLM
License	other
Context Length	131072
Model Max Length	131072
Transformers Version	4.55.4
Vocabulary Size	163840
Torch Data Type	bfloat16

Rank the K2 Merged 3.5T Fp8 capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to K2 Merged 3.5T Fp8

Best Alternatives	Context / RAM	Downloads	Likes
Kimi K2 Thinking NVFP4	256K / 210 GB	15569	29
Kimi K2 Thinking BF16	256K / 220 GB	63	6
Kimi K2 Instruct 0905	256K / 667.4 GB	17	7
DeepSeek R1 0528	160K / 176.1 GB	48	15
DeepSeek R1 0528 FP4	160K / 107 GB	72991	36
DeepSeek V3.0324 FP4	160K / 242.8 GB	9984	12
DeepSeek R1 FP4	160K / 232 GB	6856	264
DeepSeek V3.1 MLX 5.5bit	160K / GB	400	5
...V3.1 Base Int4 Mixed AutoRound	160K / 220 GB	103	2
...Seek V3.1 Int4 Mixed AutoRound	160K / 225 GB	53	11

Note: green Score (e.g. "73.2") means that the model is better than NousResearch/k2-merged-3.5T-fp8.