DeepSeek V4 Pro API pricing
As of 2026-10-31, DeepSeek V4 Pro API costs $1.445 per 1M input and $4.335 per 1M output tokens on one-api.ai (source: Cloudflare Workers AI pricing list price, checked 2026-10-22, plus our posted margin).
Early access: billing is not open yet. Prices below are what a call will debit when it opens. Request access →
DeepSeek V4 Pro is the larger DeepSeek model on one-api.ai, for coding help, multi-step reasoning and longer answers where Flash falls short. Model id: @cf/deepseek-ai/deepseek-v4-pro-0813.
Price comparison
Cloudflare's list price is what the provider charges through Cloudflare; one-api.ai is what a call debits here. one-api.ai runs DeepSeek V4 Pro on Cloudflare Workers AI, so the maker's own API price does not apply and is not shown.
Per 1M tokens
| Price | Input | Cached input | Output |
|---|---|---|---|
| Cloudflare list | $1.320 | $0.044 | $3.960 |
| hiapi | $1.50 | not posted | $4.50 |
| one-api.ai | $1.445 | $0.048 | $4.335 |
Sources: Cloudflare Workers AI pricing (page updated 2026-10-20), checked 2026-10-22; hiapi's public price page, collected 2026-10-21.
What a month costs
Three example workloads. "Monthly cost" multiplies the tokens by the one-api.ai price per 1M, and that is what billing debits: usage is metered to a millionth of a credit, with no rounding up per call. A credit is a tenth of a cent.
| Workload | Requests | Tokens in / out per request | Monthly cost | Per request |
|---|---|---|---|---|
| Support chatbot | 10,000 | 1,500 / 300 | $34.68 | 3.468 credits |
| RAG answers | 5,000 | 6,000 / 400 | $52.02 | 10.404 credits |
| Batch summaries | 1,000 | 20,000 / 1,000 | $33.235 | 33.235 credits |
Prompts that reuse a long prefix can cost less where the model has a cached-input rate, shown in the price comparison above.
Context and cache tiers
The context window is not stated on the source page, so it is not given here. There is one price tier for every prompt length. Cached input is billed at the cached rate only when the provider reports a cache hit; the cost examples above assume no cache hits. one-api.ai sets max_tokens to 4,096 when a request leaves it out and clamps it to 16,384.
Request examples
Python
import os
from openai import OpenAI
client = OpenAI(base_url="https://one-api.ai/v1/launch", api_key=os.environ["ONE_API_TOKEN"])
resp = client.chat.completions.create(
model="@cf/deepseek-ai/deepseek-v4-pro-0813",
messages=[{"role": "user", "content": "Summarise RDAP in one sentence."}],
max_tokens=512,
)
print(resp.choices[0].message.content)
print(resp.usage)JavaScript
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://one-api.ai/v1/launch", apiKey: process.env.ONE_API_TOKEN });
const resp = await client.chat.completions.create({
model: "@cf/deepseek-ai/deepseek-v4-pro-0813",
messages: [{ role: "user", content: "Summarise RDAP in one sentence." }],
max_tokens: 512,
});
console.log(resp.choices[0].message.content, resp.usage);curl
curl https://one-api.ai/v1/launch/chat/completions \
-H "Authorization: Bearer $ONE_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{"model":"@cf/deepseek-ai/deepseek-v4-pro-0813","messages":[{"role":"user","content":"Say hello in one sentence."}],"max_tokens":512}'Notes on DeepSeek V4 Pro
The id is @cf/deepseek-ai/deepseek-v4-pro-0813 on Cloudflare Workers AI. Like Flash it is subject to Cloudflare's 50 requests per minute ceiling for models paid with prepaid AI Gateway credits.
Output tokens cost three times input tokens, so long answers dominate the bill. Setting max_tokens lowers the amount reserved before a call and caps the worst case; one-api.ai defaults it to 4,096 and clamps anything above 16,384.
Our margin on V4 Pro is the same 4% over cost as Flash. Per million tokens it still lands below the price hiapi posts for this model, as the comparison table shows.
Cloudflare does not publish a context window for this model on the pricing page we cite, so none is shown. Reasoning-style answers can be long; budget output tokens accordingly.
FAQ
When should I pick V4 Pro over V4.1 Flash?
When answer quality on code or multi-step questions matters more than price. For classification, routing and short replies Flash is usually enough at a fraction of the cost.
Can I send the OpenAI messages format?
Yes. The endpoint takes an OpenAI-style chat completions body with model, messages and optional max_tokens and stream.
What happens if my balance cannot cover a call?
The call is refused with HTTP 402 before anything is sent upstream. The reservation is the worst case for that request, based on its size and max_tokens.
Related
Last updated 2026-10-31.