Delivery
One managed key for both models
Choose between DeepSeek V4 Flash and Pro, understand the shared token-credit limits, and copy the correct setup for your coding agent.
Available while stock lasts
Pay once for 30 days of managed access. Flash and Pro share the same exact token-credit windows, and your encrypted Key, Base URL, model selector, and Agent configurations appear in Dashboard.
Delivery
One managed key for both models
5-hour limit
25M Flash credits
Long-window limit
150M / 7 days
Access term
30 days · one-time
1 · How to choose
Both routes share one credit pool. Flash charges one credit per raw input/output token. Pro charges ceil(tokens × 87 / 28), or about 3.11×, so a 25M five-hour pool represents about 8.05M raw Pro tokens if every request uses Pro.
deepseek-v4-flashUse for everyday coding, repository exploration, chat, repeated tool calls, and higher-volume agent work.
deepseek-v4-proSwitch to Pro for difficult reasoning, architecture decisions, deep reviews, and final-pass verification where quality matters more than token volume.
2 · Switch models
Start with deepseek-v4-flash. For a hard reasoning or review pass, replace the request's model value with deepseek-v4-pro. Switch back to Flash for routine tool loops so the shared credit pool lasts longer.
// Default
model: "deepseek-v4-flash"// Hard task
model: "deepseek-v4-pro"3–4 · Protocol, test, then Agent
Select the exact protocol your client sends. Each option below has a different endpoint, request body, and response shape, followed by a matching direct test and Agent configuration.
These examples already use the managed customer gateway athttps://api.deepseekv4pro.com/v1. Replace only YOUR-TOKEN-PLAN-KEY with the virtual Key delivered in Dashboard. api.deepseek.com is the upstream provider endpoint and does not accept this Token Plan virtual Key.
Request → response contract
input request → output items and Responses events
Compatible tools
Do not mix /v1/responses, /v1/chat/completions, and /v1/messages. The Key and model aliases are shared; the wire contracts are not.
Direct connection test
Expected marker: responses-ok
curl https://api.deepseekv4pro.com/v1/responses \
-H "Authorization: Bearer YOUR-TOKEN-PLAN-KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4-flash",
"input": "Reply with: responses-ok",
"max_output_tokens": 32
}'Agent configuration for this protocol
Codex
~/.codex/config.toml
Install
npm install -g @openai/codex
Launch
codex
model = "deepseek-v4-flash"
model_provider = "deepseek-token-plan"
[model_providers.deepseek-token-plan]
name = "DeepSeek Token Plan"
base_url = "https://api.deepseekv4pro.com/v1"
env_key = "DEEPSEEK_TOKEN_PLAN_KEY"
wire_api = "responses"
# Put this in the shell that launches Codex:
# export DEEPSEEK_TOKEN_PLAN_KEY="YOUR-TOKEN-PLAN-KEY"Switch model
Change only the top-level model value to deepseek-v4-flash or deepseek-v4-pro, then start a new Codex task.
Verify
Run codex with a small read-only prompt and confirm the request reaches the Responses route without an authentication or model error.
Requires an OpenAI Responses-compatible gateway route. This Token Plan gateway exposes /v1/responses for that purpose.
Official client documentationPurchase strategy
Check Flash vs Pro consumption and confirm the 30-day term.
Pay $29.90 once through Stripe for a 30-day access term while inventory remains.
The encrypted virtual key and Base URL are assigned in Dashboard after payment.
Choose the model and Agent tab, copy the generated config, then run the verification prompt.