DeepSeek V4-Pro

Pricing — 1 offering(s)

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

Capability profile

Operator guidance

Route here when you want near-frontier reasoning/coding at open-model cost and V4-Flash isn't strong enough. Against proprietary frontier models (Claude Opus, GPT-5.x), V4-Pro is competitive on the cited coding/reasoning benchmarks at a fraction of the price, with the tradeoff of running a 1.6T-parameter model (self-host) or trusting DeepSeek's hosted API. For pure cost efficiency at long context, V4-Flash.

Use cases

• Frontier-adjacent open-weights reasoning and agentic coding where V4-Flash's ceiling isn't enough and cost still matters
• Very-long-context (up to 1M) analytical and coding tasks at the premium open tier
• Self-hosted deployment under MIT for teams that need a top-tier open model in their own environment

Limitations

• Benchmark figures are for the 'V4-Pro-Max' max-effort config and come from independent/secondary coverage — not a verified reading of DeepSeek's tech-report tables (SCO-164 caveat)
• Active-parameter count and architecture details are from independent coverage, not DeepSeek's own docs
• GA-date and snapshot naming vary across sources by a few weeks
• Announced peak/off-peak pricing not yet live; not reflected in the registry

Citations

• DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence (arXiv:2606.19348)
• deepseek-ai/DeepSeek-V4-Flash model card (V4 family, Hugging Face)
• DeepSeek API pricing