Whack a mole game created with a locally served LLM with LM Studio with Qwen 3.8 35B A3B MoE model by empero-ai lab.

Created in 4 minutes with 20 tokens per second in a GPU with an 8GB of VRAM. The harness is LM Studio Bionic

Husni MarwanNov 5

The part of Cursor I'd actually pay for is now open source.

OpenCode is a coding agent with more than 200,000 stars on GitHub. If you've used the agent in Cursor or Claude Code, you'll feel right at home. It reads your codebase, edits the files and runs the commands. The

Simon HøibergNov 535

This is qwen 3.8-27b result

WëscheNov 5

Ripple 65B - 60 toks/s - NVFP4

Vs Qwen 3.8 Next (122B) - 200 toks/s - NVFP4 RTX 6000 Pro No contest.

am.willNov 5144