The open coding-model race has been missing something simple: hardware reality. Poolside released Apache 2.0 weights for the 33B-total, 3B-active Laguna XS.2 and made its 225B Laguna M.1 public, betting developers want agentic coding models they can actually deploy.
LocalLLaMA reacted like dense models had suddenly become fun again. The official Qwen numbers were strong, but the real community energy came from people immediately asking about quants, GGUF builds, and whether 27B had become the practical sweet spot. By crawl time on April 25, 2026, the thread had 1,688 points and 603 comments.
Why it matters: an open-weight 27B dense model is now being pitched against much larger coding systems on real agent tasks. Qwen’s own model card lists SWE-bench Verified at 77.2 for Qwen3.6-27B versus 76.2 for Qwen3.5-397B-A17B, with Apache 2.0 licensing.
Ollama said on March 18, 2026 that MiniMax-M2.7 was available through its cloud path and could be launched from Claude Code and OpenClaw. The Ollama library page describes the M2-series model as a coding- and productivity-focused system with strong results on SWE-Pro, VIBE-Pro, Terminal Bench 2, GDPval-AA, and Toolathon.