Fine-tuned LFM2.5-2.6B (Liquid AI) for tool calling inside Odoo (ERP) via MCP. Ask in natural language in the Odoo chat and the model picks the right Odoo tool. Current version: v18 (2026-08-19) — the definitive model of the Medium line.

Part of the OdooClaw collection. The MLX release (Apple Silicon) is odooclaw-medium-2.6b-ft-mlx.

Why this model

The Medium is the agentic sweet spot of the OdooClaw family:

LFM2.5-2.6B (this model, fine-tuned): the "on-device agentic" model of the LFM2.5 family — reasons before every tool call
Light 1.2B (odooclaw-light-1.2b-ft): faster and lighter, but lower accuracy on business/finance tasks
Medium 2.6B (this model): 94.2% conversation, 99.5% creation on 1000-case batteries — the most balanced model of the series

We deliberately chose the 2.6B for agentic workloads where the model reasons before every tool call — the extra accuracy is worth the small latency cost.

⭐ Evaluation results (v18, 1000-case batteries)

Battery (1000)	v18
Conversation (990)	94.2%
Creation (1000)	99.5%
Business (1000)	74.0%
Invoices (1000)	71.6%

The v18 is the most balanced model of the series: top-2 in all 4 categories at once, no tradeoffs. Trained with balanced distribution (matches evaluation) and natural variety.

🚀 Hardware performance (v18, Q4_K_M)

Mac Mini M4 (MLX 4-bit)

Test	Time	Speed
Tool call (find partner)	0.98s	—
Long text (200 tok)	3.24s	61.7 tok/s

📦 Formats

odooclaw-medium-2.6b-ft-Q4_K_M.gguf odooclaw-medium-2.6b-ft-mlx (Apple Silicon, 4-bit) Modelfile

🔧 Quick deploy

bash

# llama.cpp
llama-server -m odooclaw-medium-2.6b-ft-Q4_K_M.gguf \
  --chat-template-file chat-template-medium-no-think.jinja \
  --temp 0.0 --jinja

# Ollama
ollama create odooclaw-medium -f Modelfile

📚 Training

LiquidAI/LFM2.5-2.6B train_lfm25_26b.py

Chat template

4-bit

License

Apache 2.0 — free for everyone, that's the whole point.

nicolasramos/odooclaw-medium-2.6b-ft

OdooClaw Medium 2.6B FT