Deepseek V3.1 NVFP4

Open weights

Deepseek V3.1 NVFP4 is a language model from DeepSeek with a 128K-token context window. Provider list price is $0.60 per million input tokens and $1.70 per million output; on Allocate you pay $0.64 and $1.82 with the 7% transaction fee. The weights are open under MIT, so you can fine-tune it and own the result.

Route Deepseek V3.1 NVFP4 Talk to sales

Pricing

Provider listOn Allocate

Input, per M tokens$0.60$0.64

Output, per M tokens$1.70$1.82

Token usage bills at the provider list price plus the 7% transaction fee. Prices checked 2026-07-08.

Price against its peers

GLM 4.7 FP8$0.45$2

Deepseek V3.1 NVFP4$0.60$1.70

Meta Llama 3.3 70B Instruct Turbo$1.04$1.04

Qwen2-VL (72B) Instruct$1.20$1.20

Qwen2.5 72B Instruct$1.20$1.20

Qwen2.5 72B Instruct Turbo$1.20$1.20

Qwen QwQ-32B$1.20$1.20

Cogito v2.1 671B$1.25$1.25

InputOutput

Provider list prices per M tokens, Deepseek V3.1 NVFP4 against its nearest language peers by price.

What a real workload costs

Take 1,000,000 requests a month at 1,200 input and 350 output tokens each: 1,200M input and 350M output tokens. At list prices that is 1,200 × $0.60 + 350 × $1.70 = $1,315 a month. Billed on Allocate it is $1,407 with the 7% transaction fee.

Deepseek V3.1 NVFP4 is an open-weights model under the MIT license. Fine-tune it on your own data and the weights stay inside your boundary; they belong to you.

Example usage

Point a route at deepseek/deepseek-v3.1 and the endpoint never changes; swap the model behind it whenever you want.

api.allocate.network

curl https://api.allocate.network/v1/chat/completions \
  -H "Authorization: Bearer $ALLOCATE_KEY" \
  -d '{
    "model": "deepseek/deepseek-v3.1",
    "messages": [{"role": "user",
      "content": "Summarise the attached contract."}]
  }'

200 · deepseek/deepseek-v3.1 · inside your boundary

Common questions

How much does Deepseek V3.1 NVFP4 cost per million tokens?

Provider list price is $0.60 per million input tokens and $1.70 per million output tokens. On Allocate you pay list plus the 7% transaction fee: $0.64 in and $1.82 out.

What context window does Deepseek V3.1 NVFP4 have?

131,072 tokens (128K). At roughly 0.75 words per token, that is about 98k words of English text per request.

Can I fine-tune Deepseek V3.1 NVFP4?

Yes. Deepseek V3.1 NVFP4 is an open-weights model under the MIT license. The license is permissive, so the fine-tuned weights are yours to use commercially. On Allocate the trained weights stay inside your boundary and belong to you.

How do I call Deepseek V3.1 NVFP4 on Allocate?

Send deepseek/deepseek-v3.1 in the model field of the OpenAI-compatible endpoint at api.allocate.network/v1, or point a route name (like prod/support-agent) at it so you can swap the model later without a deploy.