# Mlabonne releases LFM2.5-230M-Chess, a 230M-parameter chess engine language model

Digest AI · Generative AI & Models · published 2026-09-16T04:53:29Z

Canonical: https://digestai.news/story/mlabonne-releases-lfm2-5-230m-chess-a-230m-parameter-chess-engine-language-model

## Summary

A new open‑weight model, LFM2.5-230M-Chess, has been added to Hugging Face by the mlabonne community. The 230‑million‑parameter causal LM is distilled from Stockfish and treats every legal chess move as a distinct token. It reads an 80‑token board representation, predicts a win probability, then selects a move token, achieving roughly 2004 Elo with a shallow depth‑3 search and about 1500 Elo in a single‑pass policy.

The post provides step‑by‑step usage guides for popular inference stacks: Transformers pipelines, direct AutoModel loading, vLLM, SGLang, and Docker Model Runner. Sample curl commands illustrate OpenAI‑compatible API calls, and a browser demo is available via ChessLFM Space. Token‑level details, such as the 2,106 added token IDs for move encoding, are also shared.

By exposing a compact, legally‑constrained chess engine through standard AI serving tools, the model lowers the barrier for developers to embed strong chess play into applications, research, or educational tools without needing heavyweight search engines.

## Key points

- LFM2.5-230M-Chess is a 230M‑parameter LM fine‑tuned via Stockfish distillation, playing ~2004 Elo with depth‑3 search.
- Every legal chess move is a separate token; the model reads an 80‑token FEN prompt and outputs win probability then move.
- Full integration guides are provided for Transformers, vLLM, SGLang, and Docker, with OpenAI‑compatible API examples.

## Why it matters

The model democratizes access to strong, legally‑constrained chess AI, enabling lightweight integration into apps and research without costly compute.

## Sources

1. [mlabonne/LFM2.5-230M-Chess · Hugging Face](https://huggingface.co/mlabonne/LFM2.5-230M-Chess) (huggingface.co, 2026-09-16, primary source)

## Cite

Digest AI, "Mlabonne releases LFM2.5-230M-Chess, a 230M-parameter chess engine language model", 16 September 2026, https://digestai.news/story/mlabonne-releases-lfm2-5-230m-chess-a-230m-parameter-chess-engine-language-model

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/mlabonne-releases-lfm2-5-230m-chess-a-230m-parameter-chess-engine-language-model.json