JetBrains ships Mellum2.1 open coding model with RL-heavy post-training; llama.cpp MTP support submitted
jacek2023 · reddit · 2026-10-16
JetBrains released Mellum2.1 Thinking and a llama.cpp pull request (#30612) already adds MTP support for it.
- Architecture unchanged: 12B-total MoE with 2.5B active params, 64 experts (8 activated), 131,072 context, Apache 2.0
- This version's work went almost entirely into post-training: RL became the main training phase rather than a short final stage, with new RL tasks in math, competitive programming, science, tool use and software engineering, all sources filtered before training
- Agentic skills trained in real repos: the model works with a shell and file-editing tools, rewarded when tests pass
- After millions of sandboxed runs, JetBrains says it explores codebases, edits files and verifies its own changes, clearly outperforming Mellum2 on agentic tasks
Related event: JetBrains Releases Open-Source Thinking Model Mellum 2.1 on Hugging Face(2 posts)→
More from coding & agent
- Flux.ai imports STEP files and edits existing CAD geometry via natural language — burhop · 2026-10-16
- 50 tasks at once energizes me, says dev, as peer warns AI context-switching causes burnout — zack_overflow · 2026-10-16
- Founder wipes 30k lines of tech debt in one shot with Opus — lucasmeijer · 2026-10-16
- Japanese engineer runs an AI agent org inside tmux: per-project teams, one human talking to the orchestrator — 4310sy · 2026-10-16
- gloria-mcp ships an MCP server serving Gloria AI's curated crypto news — modelcontextprotocol · 2026-10-16
- Pine Script MCP server lets AI agents read v6 docs and validate syntax — modelcontextprotocol · 2026-10-16