DeepSeek V4 API without a DeepSeek account: pay per token

DeepSeek's V4 line is one of the strongest price-to-quality LLM families available right now, but using it directly means a DeepSeek account, a card its billing will accept, and a separate balance from every other model you already pay for. genmux puts deepseek-v4-pro, deepseek-v4-flash and deepseek-v3 behind the same API key as every other text and image model on the platform — one balance, one Authorization header, no separate signup.

Pro, Flash or V3

All three are OpenAI-compatible chat models reachable at POST /v1/chat/completions. Pick by task, not by habit:

Both figures are what genmux bills, converted from credits at $0.0125/credit (the Scale pack rate, 12,000 credits for $150). Flash and V3 are priced the same per token; the difference is context window and whether thinking is available.

Calling it with curl

curl https://api.genmux.tech/v1/chat/completions \
  -H "Authorization: Bearer $GENMUX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [
      {"role": "system", "content": "You are a concise support agent."},
      {"role": "user", "content": "Summarize this ticket in two sentences: ..."}
    ],
    "max_tokens": 300,
    "temperature": 0.3
  }'

The response is a standard chat.completion object. The id field is the genmux job id — save it if you want to look the call up later with GET /v1/generations/{id}, or list every chat call your account has made with GET /v1/generations?capability=text.chat.

Reading the real cost from usage

from openai import OpenAI

client = OpenAI(base_url="https://api.genmux.tech/v1", api_key="nb_live_...")

resp = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Draft a refund policy for a SaaS product."}],
    max_tokens=600,
)

print(resp.choices[0].message.content)
print(resp.usage)
# Usage(prompt_tokens=41, completion_tokens=387, total_tokens=428,
#       cost=0.523, cost_source="vendor")

usage.cost is the exact number of credits deducted for this call; cost_source tells you whether that came straight from the upstream vendor (vendor) or was computed from token counts (estimated) — streaming responses always fall in the second bucket, since DeepSeek doesn't report usage for streamed output. Before the call runs, genmux reserves credits for your prompt plus max_tokens (capped at the model's own default) and settles the difference afterward, so a short reply never gets billed at the reserved ceiling.

Cheaper than DeepSeek's own list price

Our pricing rule is simple: we sit 10% under the model owner's own list price wherever our supply cost allows, and never below it. DeepSeek's own off-peak list price, checked 2026-09-06:

DeepSeek doubles its own price during peak hours (01:00–04:00 and 06:00–10:00 UTC on weekdays) — genmux doesn't vary by time of day, so during DeepSeek's peak the gap widens to about 55% cheaper on both input and output.

Streaming and thinking

Both are supported the same way as with OpenAI's own API: pass stream: true to get server-sent events, with the final chunk before data: [DONE] carrying the usage object (see our streaming guide for the exact parsing code). On V4 Pro and V4 Flash, add "enable_thinking": true to let the model reason before answering — reasoning tokens are billed as output tokens, so check how thinking is billed before turning it on for high-volume traffic.

Getting started

Payments today are crypto (USDT); card and SBP payments are being enabled. Grab a key from the dashboard, point the OpenAI SDK's base_url at https://api.genmux.tech/v1, and you can be calling deepseek-v4-flash in the time it takes to read the docs. Model details and live rates live on the DeepSeek V4 Pro and DeepSeek V4 Flash pages, and the full lineup is on the pricing page.