DeepSeek
DeepSeek V4 Pro
The open-weights flagship — AA index 52 at one-tenth of flagship prices.
Open weightstextREL 2026-04-24
Compare this model →Verified 2026-06-12
Overview
DeepSeek V4 Pro is the strongest open-weights model available: Intelligence Index 52 (Artificial Analysis) — within striking distance of closed flagships — at $0.43/$0.87 per 1M tokens via API, or free on your own hardware. 1M context, 384K max output, MIT-style licensing.
Benchmark scores
Methodology →Intelligence52
Codingn/a
Reasoningn/a
Mathn/a
Sources and verification
- Intelligence: Artificial Analysis Intelligence Index · checked 2026-06-12
- Catalogue and pricing: OpenRouter model record · checked 2026-06-12
Strengths and weaknesses
Strengths
- Best open-weights model
- ~10× cheaper than closed flagships
- 384K output tokens
- Self-hostable
Trade-offs
- Text-only
- Needs serious hardware to self-host (multi-GPU)
Reach for this model when
Anyone wanting near-frontier quality with open weights — via API or self-hosted.
Run something like it locally
This IS the local option — run it yourself with vLLM on a multi-GPU server, or quantised via llama.cpp on high-VRAM workstations.
