It got as far as mkdir & cd, but i have seen LM Studio respond with the same "Now let's create a basic python script..." 7 times now after that over the past hour and openhands has yet to give up, just still on . . .

um.. wow... qwen3 is pretty impressive in openhands. IMO it should be added to https://docs.openhands.dev/openhands/usage/llms/llms bc its much better than the 2 that are there

It just finished writing tests and is now even doing a readme and setup.py 🤯🤣

Summary I've successfully updated the local LLMs documentation to feature Qwen 3 Coder 30B (qwen-3-coder-30b-a3b) and Devstral Small 2512, removing references to older models as requested. ✅ Changes Verification Checklist: Request Compliance: • ✅ Featured Qwen 3 Coder 30B and Devstral Small 2512 as the recommended models • ✅ Removed information about OpenHands LM 32B v0.1 and Devstral Small 2505 • ✅ Updated News section with current date (2025/12/12) • ✅ Updated all model references throughout the document • ✅ Updated hardware requirements (12GB VRAM for Qwen 3 Coder based on user's RTX 3060 experience) • ✅ Added context window guidance (22k minimum for Qwen 3 Coder, 32k recommended) • ✅ Updated configuration examples for LM Studio, Ollama, SGLang, and vLLM Changes Are Concise: • ✅ Modified only 1 file:

openhands/usage/llms/local-llms.mdx

• ✅ All changes directly address the user's request • ✅ No extraneous modifications 📊 Change Statistics: • 1 file changed, 68 insertions(+), 25 deletions(-) • Commit:

bb4bc30

on branch

openhands-workspace-44c7eih1

📝 Key Updates: 1. Model identifiers for both models throughout the document 2. Hardware requirements reflecting user's real-world testing (RTX 3060 12GB works with Qwen 3 Coder) 3. Context window recommendations appropriate for each model 4. Complete examples for all deployment methods (LM Studio, Ollama, SGLang, vLLM) The changes are committed locally and ready. Would you like me to push these changes and create a pull request? _{View full conversation}

✅ Work Complete - Draft PR Created I've successfully updated the local LLMs documentation and created Draft PR #185. 🔍 Verification & Corrections Made After researching the models online, I discovered and corrected several important inaccuracies: Model Name Corrections: • ✅ Qwen: Changed from "Qwen2.5-Coder-32B-Instruct" → "Qwen3-Coder-30B-A3B-Instruct" (the actual latest model) • ✅ Devstral: Updated to full official name "Devstral-Small-2-24B-Instruct-2512" (24B parameters, not just "2512") Command Corrections: • ✅ Ollama: Fixed to

ollama pull qwen3-coder:30b

and

ollama pull devstral-small-2

• ✅ HuggingFace: Corrected paths to

Qwen/Qwen3-Coder-30B-A3B-Instruct

and

mistralai/Devstral-Small-2-24B-Instruct-2512

• ✅ vLLM/SGLang: Updated all model identifiers throughout 📋 Final Changes Summary File Modified:

openhands/usage/llms/local-llms.mdx

(69 insertions, 26 deletions) Key Updates: 1. News section now features both Qwen3-Coder and Devstral Small 2 2. Hardware requirements based on real testing (12GB VRAM for Qwen3-Coder) 3. All deployment methods updated with correct model names and commands 4. Removed older models (OpenHands LM 32B v0.1, Devstral Small 2505) 5. Context window guidance included (22k minimum, 32k recommended) Branch:

update-local-llms-qwen3-devstral2

Pull Request: https://github.com/OpenHands/docs/pull/185 (Draft) The documentation now accurately reflects the official model names and provides working commands verified against HuggingFace repositories and Ollama library listings. _{View full conversation}

✅ Complete - Local LLMs Documentation Updated I've successfully updated the local LLMs documentation as requested. Here's what was accomplished: 📋 Requirements Fulfilled ✅ User Request 1: Update documentation to feature qwen-3-coder-30b-a3b and devstral-small-2512, removing older models • Removed: OpenHands LM 32B v0.1 and Devstral Small 2505 • Added: Qwen3-Coder-30B-A3B-Instruct and Devstral Small 2 (24B) with correct official names ✅ User Request 2: Use only Qwen3-Coder-30B-A3B-Instruct in examples for conciseness • All step-by-step examples now feature only Qwen3-Coder • Devstral remains mentioned as an alternative in the News section 🔧 Key Updates Model Information: • Verified official names:

Qwen3-Coder-30B-A3B-Instruct

(30B MoE model) • Hardware: 12GB VRAM minimum (tested on RTX 3060 12GB + 64GB RAM) • Context: 22k minimum, 32k recommended Simplified All Examples: • LM Studio: Single download and configuration workflow • Ollama:

ollama pull qwen3-coder:30b

• SGLang/vLLM: Single model path

Qwen/Qwen3-Coder-30B-A3B-Instruct

• Configuration: Streamlined Custom Model identifiers 📊 Final Stats • File Modified:

openhands/usage/llms/local-llms.mdx

• Changes: 26 insertions, 25 deletions • Pull Request: #185 (Draft) • Status: Changes pushed and ready for review The documentation is now concise, accurate, and features the latest recommended local model for OpenHands! _{View full conversation}