大模型网摘V4-Pro-0813 API Access and Pricing: Official $0.66/$1.98, a 52-Provider Market Survey
V4-Pro-0813 is one of the most widely carried flagship snapshots in DeepSeek's ecosystem: beyond the official API, aggregators list 52 providers serving the model. Same model, different channels — price gaps up to 3x. This guide turns the official and representative third-party numbers into a quick-reference table so you can pick a channel by workload.
The official endpoint: the baseline
The official API exposes the model as deepseek-v4-pro at $0.66 input / $1.98 output per million tokens (as of press time), with 1,048,576 context, a 393,216 output cap, and per-request thinking/non-thinking switching. Batch and off-peak discounts apply here too — the off-peak mechanism is covered in our off-peak pricing guide, and the model itself in our V4-Pro-0813 deep dive.
Third-party landscape: what convenience costs
Representative channels per models.dev aggregation (as of press time):
| Provider | In $/M | Out $/M | Notes |
|---|---|---|---|
| DeepSeek official | 0.66 | 1.98 | Baseline; 393K output cap |
| AIHubMix | 0.69 | 2.08 | Aggregator, near-official |
| above.dev | 0.73 | 2.18 | Aggregator |
| Deep Infra | 1.30 | 2.60 | Mainstream hosting, 384K out |
| Cloudflare / Together / Baseten | 1.32 | 3.96 | Cloud tier, consolidated billing |
| Cortecs | 2.00 | 3.99 | Only 64K output cap — careful with long tasks |
The market has a two-tier structure: aggregators track the official price (a 5–10% premium) while mainstream clouds charge roughly double. Watch the context and output caps — the same model ID ships with 384K output on some providers and as little as 64K on others. Verify before launching long-output jobs.
A worked example: a typical agent task
Suppose a code-comprehension job reads a 200K-token document and writes 10K tokens back, run 100 times. On the official API: 20M×$0.66 = $13.2 input, 1M×$1.98 = $1.98 output — about $15.2 total. On a $1.32/$3.96 cloud tier: about $30.4 — exactly double. At scale, that gap covers the cost of migrating back to the official API, as laid out in our migration checklist.
Selection advice
Three rules of thumb: volume decides the channel — beyond a million tokens a day, prefer the official API or budget aggregators; during experiments, whatever cloud account you already have is fine. Check context and output caps before committing to avoid mid-task truncation. And watch the billing model — some platforms bill per request rather than per token, which can backfire on long-context workloads. If you also need live web data, pair the model with the Search & Fetch API — see our endpoints and billing guide.