DeepSeek AI

Overview

DeepSeek’s API is deepseek-flash as of the Sep 2026 docs fetch (api-docs.deepseek.com). The previous deepseek-v4-flash and deepseek-v4-flash-vision-exp model names are still accepted as aliases — DeepSeek retires them as model names but routes their requests to V4.1 Flash at the Flash price.

The other advertised model is deepseek-v4-pro — being retired. Per DeepSeek’s own announcement: “V4.1 Flash has comprehensively surpassed V4 Pro in performance, cost, speed, and total time.” From 12:00 Beijing Time on September 14, 2026 until V4.1 Pro is released, requests to deepseek-v4-pro are routed to V4.1 Flash and billed at the Flash price.

The API is OpenAI/Anthropic-compatible (base_url=https://api.deepseek.com for OpenAI SDK, …/anthropic for Anthropic SDK). Models support thinking: { type: "enabled" } and reasoning_effort — call them like reasoning models, not chat models. DeepSeek Harness is in developer preview for agent-harness developers.

Key Strengths

API pricing — pick a route

Per 1M tokens, USD. DeepSeek’s own first-page text extract does not include a USD price table (their docs are a JS app); use api-docs.deepseek.com for the live USD table. Vendor-routed prices below came from each vendor’s live pricing page this session.

Vendor-hosted DeepSeek V4 Flash 0731 is the cheapest routable model in this set. Until V4.1 Pro ships, V4 Pro is effectively a Flash-priced alias.

⚠️ Privacy: Access via US Vendors

Critical: DeepSeek’s own API and web chat route data through Chinese servers. For any business or professional use, access DeepSeek weights through US-based inference providers.

→ Complete Privacy Guide — How to use DeepSeek safely

Recommended US providers

Quick start (Together, OpenAI-compatible)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.together.xyz/v1",
    api_key="your-together-key",
)

resp = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4-Pro-0813",
    messages=[{"role": "user", "content": "Refactor this module…"}],
    extra_body={"thinking": {"type": "enabled"}, "reasoning_effort": "high"},
)
print(resp.choices[0].message.content)

Or hit DeepSeek’s own endpoint with the new deepseek-flash name:

curl https://api.deepseek.com/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${DEEPSEEK_API_KEY}" \
  -d '{"model": "deepseek-flash",
       "messages": [{"role":"user","content":"Hello!"}],
       "thinking": {"type": "enabled"},
       "reasoning_effort": "high",
       "stream": false}'

Agent integrations

From api-docs.deepseek.com, DeepSeek is supported directly as a backend model by Claude Code, GitHub Copilot, and OpenCode — no code changes needed. Just point them at a Together or Fireworks endpoint with the model id.

DeepSeek for Your Industry

Comparisons

• Best models by lab
• DeepSeek vs ChatGPT — Cost vs ecosystem
• DeepSeek vs Claude — Budget vs premium
• Local LLM vs Cloud — Self-hosted DeepSeek options