DeepSeek API
DeepSeek is a first-party model API with OpenAI-compatible and Anthropic-compatible endpoints. The 2026 V4-Pro pricing cut matters for Kevin's stack because it changes model-routing economics, but it still needs evals, provenance, and spend controls before agent workloads move.
Current API Shape
DeepSeek's docs list first-party base URLs for both OpenAI-format and Anthropic-format clients:
| Client format | Base URL |
|---|---|
| OpenAI-compatible | https://api.deepseek.com |
| Anthropic-compatible | https://api.deepseek.com/anthropic |
The current model table exposes deepseek-v4-flash and deepseek-v4-pro, both with thinking mode, JSON output, tool calls, chat prefix completion, and beta FIM completion in non-thinking mode. The docs also say deepseek-chat and deepseek-reasoner map to V4-Flash non-thinking/thinking modes and will be deprecated on 2026-07-24 15:59 UTC. Source: DeepSeek API docs, 2026-07-03
Pricing Snapshot
Deep review on 2026-07-03 confirmed the bookmark image against the official pricing page:
| Model | Input cache hit | Input cache miss | Output | Context | Max output | Concurrency |
|---|---|---|---|---|---|---|
deepseek-v4-flash |
$0.0028 / 1M tokens |
$0.14 / 1M tokens |
$0.28 / 1M tokens |
1M | 384K | 2500 |
deepseek-v4-pro |
$0.003625 / 1M tokens |
$0.435 / 1M tokens |
$0.87 / 1M tokens |
1M | 384K | 500 |
The local image says the V4-Pro 75% discount is now permanent and shows the old crossed-out prices: $0.0145 cache-hit input, $1.74 cache-miss input, and $3.48 output per 1M tokens. The official docs currently show only the post-cut prices and warn that product prices may vary, so treat this as a dated source snapshot, not a forever routing guarantee. Source: X/@deepseek_ai, 2026-05-22; Source: DeepSeek pricing, 2026-07-03
Stack Fit
- Use Agent Unit Economics before moving production work: low token price is useful only if task success rate, latency, retries, data handling, and support burden hold.
- Use OpenRouter or Ai Gateway when provider fallback, central telemetry, routing policies, or app-level model abstraction matter more than direct first-party pricing.
- Use Agent Security Model for provenance checks. This is first-party pricing, unlike grey-market token resale, but it still requires API-key hygiene, data-retention review, and spend caps.
- Keep model adoption eval-gated. DeepSeek's economics make it a candidate for high-volume agent and coding workloads, not an automatic default.
Timeline
- 2026-05-22 | DeepSeek announced the V4-Pro 75% API discount would become permanent. 23,542 likes and 5,837 bookmarks in Kevin's library. Source: X/@deepseek_ai, 2026-05-22
- 2026-07-03 | Created this page after one-by-one artifact review. Verified the local pricing image against official DeepSeek docs and recorded current V4-Flash/V4-Pro prices, base URLs, context/output limits, concurrency limits, and deprecation note for
deepseek-chat/deepseek-reasoner. Source: DeepSeek API docs, 2026-07-03