大模型网摘Nine Coding Agents Now Work with the DeepSeek API: Harness, Claude Code, Codex and More
If you want to run DeepSeek models inside a coding agent, the options have multiplied. The Agent Integrations section of the official docs now lists nine products: DeepSeek Harness, Claude Code, Codex, OpenCode, OpenClaw, Hermes, Reasonix, WorkBuddy/CodeBuddy and Qoder. Combined with API endpoints that speak both the OpenAI and Anthropic formats, most mainstream coding agents can be pointed at deepseek-flash or deepseek-v4-pro with a config change.
Two integration routes: swap the base_url, or use the official Harness
Route one is a zero-refactor environment-variable play: point an OpenAI-compatible client at https://api.deepseek.com, or an Anthropic-compatible client at https://api.deepseek.com/anthropic, then add your API key and model name — Claude Code, Codex and OpenCode all work this way. Route two is DeepSeek's official Harness, a runtime framework for agent developers, now at v0.2.1. One detail worth noting: when used through Harness, Claude Code or Codex, server-side Search & Fetch tool calls are currently free of charge (per-request billing has not started, as of press time).
How to pick among the nine
By scenario: DeepSeek Harness suits developers who want the official native experience with Search & Fetch and structured outputs. For Claude Code users, the main motive to switch is cost — under peak/off-peak pricing, Flash output drops to $0.6 per million tokens off-peak. Codex is the natural hop for OpenAI-ecosystem users; open-source terminal tools like OpenCode and OpenClaw fit the self-hosting and deep-customization crowd; Reasonix, WorkBuddy/CodeBuddy and Qoder target specific workflows, each with its own setup guide in the docs.
The self-hosting route is filling in too
If you would rather skip the API entirely, DeepSeek's open-source production inference framework DeepInfer self-hosts DeepSeek-V4.1-Flash on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs, with structured outputs and deterministic sampling — enough to move a whole coding-agent stack in-house. Combined with our earlier agentic coding cost comparison, the break-even point between self-hosting and API deserves its own write-up; the Reviews channel will follow up.