DeepSeek-V4-Pro
DeepSeek-V4-Pro is DeepSeek's largest open model, a 1.6T-parameter mixture-of-experts with 49B active parameters and a 1M-token context, released under MIT. It launched as a preview in April 2026 and reached general availability as V4-Pro-0813 in August 2026 with low, high and max reasoning effort. DeepSeek has said it will be replaced by a V4.1-Pro model.
Specifications
- Context window
- 1Mabout 786,000 words, or 1,570 pages
- Max output
- 375Ktokens per response
- Input price
- $1.32per 1M tokens
- Output price
- $3.96per 1M tokens
- Cached input
- $0.044per 1M tokens
- Weights
- OpenMIT
- Parameters
- 1.6T total, 49B activeMoE
- Input types
- Text
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Apr 24, 2026Current
DeepSeek API peak price (model deepseek-v4-pro, served by V4-Pro-0813). Off-peak: $0.66 input, $0.022 cached, $1.98 output. DeepSeek's 2026-09-10 announcement says V4-Pro is being phased out ahead of a V4.1-Pro release, but its change log says V4-Pro API service continues past 2026-09-14 with billing unchanged, and the pricing page still lists it. Released as a preview on 2026-04-24; general-availability update 2026-08-13 (the 0813 checkpoint stores 1.65T parameters including a speculative-decoding module).
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $127 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $38 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $103 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- MIT-licensed weights at 1.6T scale
- 1M-token context and up to 384K output tokens
- Thinking and non-thinking modes with effort levels
Trade-offs to weigh
- Being phased out in favor of the V4.1 generation
- Self-hosting needs a large multi-GPU cluster even with FP4/FP8 weights
Where it fits
DeepSeek positions DeepSeek-V4-Pro for:
Where you can use it
- Hugging Face
- Ollama
- API
- Together AI
- DeepInfra
Run it yourself
Released under MIT. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 3,200 GB |
| 8-bit (FP8/INT8) | 1,600 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 970 GB |
This is a mixture-of-experts model: all 1.6T total parameters must sit in memory, but only about 49B are used per token, so it runs faster than a dense model of the same size.
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| GLM-5.3Z.ai (Zhipu AI) | Frontier | 1M | $1.40 / $4.40 | Compare |
| Qwen3.8-2.4T-A95BAlibaba (Qwen) | Frontier | 256K | $2 / $6 | Compare |
| Mistral Large 4Mistral AI | Frontier | 1M | $1.36 / $4.18 | Compare |
| Grok 4.7xAI | Frontier | 500K | $2 / $6 | Compare |
Frequently asked questions
How much does DeepSeek-V4-Pro cost?
DeepSeek-V4-Pro lists at $1.32 per million input tokens and $3.96 per million output tokens, with cached input at $0.044 (checked October 10, 2026). DeepSeek API peak price (model deepseek-v4-pro, served by V4-Pro-0813). Off-peak: $0.66 input, $0.022 cached, $1.98 output. DeepSeek's 2026-09-10 announcement says V4-Pro is being phased out ahead of a V4.1-Pro release, but its change log says V4-Pro API service continues past 2026-09-14 with billing unchanged, and the pricing page still lists it. Released as a preview on 2026-04-24; general-availability update 2026-08-13 (the 0813 checkpoint stores 1.65T parameters including a speculative-decoding module). Confirm current rates on the provider's pricing page before you commit.
What is the context window of DeepSeek-V4-Pro?
DeepSeek-V4-Pro accepts up to 1,048,576 tokens of context (about 786,000 words, or 1,570 pages), and it can write up to 375K tokens in a single response. Long prompts cost more and can be slower, so most apps send only what the task needs.
Is DeepSeek-V4-Pro open source?
DeepSeek-V4-Pro is an open-weight model released under the MIT license. You can download the weights from Hugging Face (deepseek-ai/DeepSeek-V4-Pro-0813) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run DeepSeek-V4-Pro locally?
Yes, if you have enough memory. The weights alone need about 970 GB at 4-bit quantization and about 3,200 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is DeepSeek-V4-Pro best for?
DeepSeek positions DeepSeek-V4-Pro for coding, agents, reasoning, long documents. It accepts text as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813
- api-docs.deepseek.com/quick_start/pricing
- api-docs.deepseek.com/news/news260424
- api-docs.deepseek.com/news/news260813
- api-docs.deepseek.com/news/news260910
- api-docs.deepseek.com/updates
- huggingface.co/deepseek-ai/DeepSeek-V4-Pro
- huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/raw/main/config.json
- huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Pro
- huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Pro-0813
- docs.together.ai/docs/serverless-models
- deepinfra.com/pricing
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.