LLM
GA
Open source
DeepSeek V4 Pro
Best price/intelligence balance; open-weight frontier reasoning
Release2026-08
Context1M
Max output131,072 tokens
Parameters1.6T
Local / VRAM1400–2000 GB
LicenseOpen source
Overview
DeepSeek V4 Pro is a model in the LLM category from DeepSeek. It was released in 2026-08. The vendor is based in China. Best for: Best price/intelligence balance; open-weight frontier reasoning.
Hardware: FP8 ~1.6 TB; multi-H100/B200 cluster.
Strengths
- 1.6T MoE
- 1M context
- MIT license
- Very cheap API
Limits
- Self-hosting is hard in practice
- No visual input
Specifications
| Vendor | DeepSeek |
|---|---|
| Category | LLM |
| Release | 2026-08 |
| Status | GA |
| License | Open source |
| License name | MIT |
| Commercial use | Allowed |
| Parameters | 1.6T |
| Active parameters | 49B |
| Architecture | MoE transformer |
| Context window | 1M |
| Max output tokens | 131,072 |
| Input modalities | Text |
| Output modalities | Text |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| VRAM | 1400–2000 GB |
| System RAM | 1600–3000 GB |
| API access | Yes |
| Pricing | $0.435 / $0.87 per 1M token |
| Confidence | High |
| Vendor country | China |
| Vendor type | Open lab |
Pricing & hardware
$0.435 / $0.87 per 1M token
This model can be self-hosted or run locally. Approximate GPU memory: 1400–2000 GB. Approximate system RAM: 1600–3000 GB. API access is available.