deepseek-flash for the current Flash model, or deepseek-v4-pro for Pro. These model IDs were checked against the provider catalog on October 24, 2026.
DeepSeek exposes an OpenAI-compatible API, so Agentium connects using the standard openai SDK pointed at https://api.deepseek.com.
DeepSeek enforces account-level concurrency limits and returns HTTP 429 when they are exceeded. See Rate limit and isolation.
Setup
- Install
- Environment
Factory
Supported models
The old
deepseek-chat and deepseek-reasoner names were retired. The deepseek-v4-flash and deepseek-v4-flash-vision-exp aliases are still accepted by DeepSeek but now resolve to the current Flash model. Agentium forwards your chosen ID unchanged and records the returned model identity separately; it does not guess a replacement or a tariff. See the current catalog and retirement notice.
Basic example
Reasoning
Setreasoning.enabled to send thinking: { type: "enabled" } and reasoning_effort. DeepSeek only accepts low, high, and max, so medium is sent as high. effort: "none" sends thinking: { type: "disabled" }. Tool turns replay reasoning_content.
The provider returns supported reasoning content in reasoning_content:
Tool calling
Full example
The tracker preserves reported usage. A missing tariff or billing field makestotal null. Supply a catalog that matches your billing provider, API, model or deployment, tier, and region; see cost accounting.