Claude code runs on DeepSeek, GLM and Qwen using two environment variables
anthropic claude deepseek qwen
| Source: Mastodon | Original article
Claude Code can now run on DeepSeek, GLM and Qwen backends using just two environment variables, communicating via the Anthropic Messages format.
Claude Code, Anthropic’s code‑generation assistant, can now be pointed at alternative large‑language‑model back‑ends such as DeepSeek, GLM and Qwen simply by setting two environment variables. The tool communicates with its back‑end using the Anthropic Messages format, meaning any endpoint that accepts that payload can serve as a drop‑in replacement for Anthropic’s own API.
The tweak opens the door for developers to run Claude Code on open‑source models without rewriting the client. By leveraging the same message schema, users can experiment with different model families, potentially lowering costs and sidestepping rate limits tied to Anthropic’s service. It also broadens the tool’s appeal to communities that prioritize model transparency and local execution, reinforcing the inclusive, community‑driven ethos that has surrounded Claude’s recent outreach efforts.
The move matters because it blurs the line between proprietary AI assistants and the growing ecosystem of open‑source LLMs. If Claude Code performs competitively on these alternative back‑ends, it could accelerate adoption of open‑source models for software development tasks, challenge the dominance of commercial APIs, and spur further integration work across the AI tooling landscape.
What to watch next: performance benchmarks comparing Claude Code’s output on DeepSeek, GLM and Qwen versus Anthropic’s own models; any official response from Anthropic regarding supported back‑ends or licensing; and whether other AI‑coding agents adopt a similar “environment‑variable” approach to broaden compatibility. The community’s reaction—particularly around security, reliability and cost implications—will shape how quickly this flexibility translates into real‑world productivity gains.
Sources
Back to AIPULSEN