deepseek
236B parameters
Commercial OK
Reviewed May 2026

DeepSeek V2.5 236B

DeepSeek V2.5 — merged V2 chat + Coder. Pre-V3 baseline; 21B active MoE.

License: DeepSeek License·Released Sep 5, 2024·Context: 131,072 tokens

Overview

DeepSeek V2.5 — merged V2 chat + Coder. Pre-V3 baseline; 21B active MoE.

Family & lineage

How this model relates to others in its lineage. Family members share architecture and training-data roots; parent / children edges record direct distillation or fine-tune relationships.

Strengths

  • Multi-head latent attention
  • 21B active MoE

Weaknesses

  • V3 / V4 supersede for new deployments

Quantization variants

Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.

QuantizationFile sizeVRAM required
Q4_K_M134.0 GB160 GB

Get the model

HuggingFace

Original weights

huggingface.co/deepseek-ai/DeepSeek-V2.5

Source repository — direct quantization required.

Hardware that runs this

Cards with enough VRAM for at least one quantization of DeepSeek V2.5 236B.

Compare alternatives

Models worth comparing

Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.

Step up
More capable — bigger memory footprint
No verdicted models in the next tier up yet.

Frequently asked

What's the minimum VRAM to run DeepSeek V2.5 236B?

160GB of VRAM is enough to run DeepSeek V2.5 236B at the Q4_K_M quantization (file size 134.0 GB). Higher-quality quantizations need more.

Can I use DeepSeek V2.5 236B commercially?

Yes — DeepSeek V2.5 236B ships under the DeepSeek License, which permits commercial use. Always read the license text before deployment.

What's the context length of DeepSeek V2.5 236B?

DeepSeek V2.5 236B supports a context window of 131,072 tokens (about 131K).

Source: huggingface.co/deepseek-ai/DeepSeek-V2.5

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.

Related — keep moving

Before you buy

Verify DeepSeek V2.5 236B runs on your specific hardware before committing money.