Autodesk ML researcher AK showed how to point Anthropic's Claude Code CLI at the open-weight GLM-5.2 model hosted on Hugging Face, letting the terminal agent run coding and agentic tasks through open inference providers instead of closed Anthropic endpoints.
I use glm 5.2 in claude code via hf claude almost daily now
moved over completely to open models
GLM-5.2 is now selectable in Claude Code via Hugging Face🤗 Inference Providers + hf-claude.
Open models are becoming easier to plug directly into real developer workflows. 😀
Developers set an HF token and environment variables so Claude Code lists GLM-5.2 among selectable models; the setup routes requests to Hugging Face Inference Providers while keeping the existing agent loop intact.
GLM-5.2 weights are public under MIT license with strong coding benchmark numbers, yet per-token pricing, quotas, and long-term availability through the hf-claude integration have not been confirmed beyond limited promotional notes.
Positive users praise Hugging Face adding GLM-5.2 to Claude Code as a game changer for fitting open models into existing workflows, while negative users report failures on complex refactors and path handling.
No Digg Deeper questions have been answered for this story yet.
Here is the reference doc with an example of using Claude Code with Hugging Face Inference Providers:
https://huggingface.co/docs/inference-providers/en/integrations/claude-code#billing-to-an-organization
GLM-5.2 is now selectable in Claude Code via Hugging Face🤗 Inference Providers + hf-claude.
Open models are becoming easier to plug directly into real developer workflows. 😀
@_akhaliq i had a bad time getting this work to with cygwin, glm52 just couldnt deal with the paths and kept going off on a tangent. it cost $7 on hf to do nothing, but it was interesting to try claude code with another model.
@StatsWire Check out: https://huggingface.co/docs/inference-providers/en/integrations/vscode
Hey kfirgollan — hf-claude is an official Hugging Face CLI extension (docs + GitHub under huggingface org). It simply sets `ANTHROPIC_BASE_URL=https://router.huggingface.co` + your HF token so Claude Code routes requests there instead of Anthropic’s API.
Anthropic explicitly supports this pattern. Their docs and enterprise setup cover custom base URLs, LLM gateways, Bedrock, Vertex AI, and proxies. It’s a built-in knob, not a hack.
No known legal battle here. HF isn’t proxying Anthropic models or scraping — they’re exposing their own Inference Providers through a compatible endpoint. Users pay HF directly. Similar setups (LiteLLM, Portkey, self-hosted gateways) are common and tolerated.
Big company or small, the risk looks low when it’s just routing to open models you choose. Clean integration on HF’s side.
@_akhaliq GLM 5.2 is genuinely better than any frontier model right now for reverse engineering idc I stand on it
@grok I'm genuinely interested in the legal aspect here. Like so far I just assumed Anthropic doesnt permit this but small companies/oss don't mind the risk and Anthropic doesnt fight it. But a big company like hugging face? Is there some legal battle here? Or maybe it's not an official HF product?
@_akhaliq How to integrate it in vs code copilot?
@StatsWire @_akhaliq Way too expensive to run it locally at its max potential lol, I just use API
@johnmccoyx @_akhaliq How to use it internally without API for security?
@johnmccoyx @_akhaliq Thanks for sharing the feedback and saving my time 😀
@zRdianjiao great news. glad to see open models moving so fast
@zRdianjiao harness compatibility is the new 'runs on linux'
@zRdianjiao Plugging open models into the same harness as frontier ones is useful for cost sensitive or air gapped work. The open question is how much the surrounding agent scaffolding still needs frontier level reasoning to stay reliable on ambiguous tasks.
@zRdianjiao GLM-5.2 plugged into Claude Code with two environment variables and a HF token.
the harness stays, the model swaps.
@_akhaliq wait, hf claude lets u plug glm directly into code workflows now? that setup wouldve saved me hours last week
@zRdianjiao Open models slotting into Claude Code's interface instead of competing with it is the more interesting trend here - the moat is shifting from "which model" to "which workflow people already trust."
@_akhaliq tried going all open, but refactors that span three files still make it hallucinate edits. i switch back for those.
@_akhaliq 这切换速度也太快了，体验稳吗
@_akhaliq having an open model that actually fits into the existing claude workflow is lowkey the game changer
