First Fine-tunes of LFM2.5 Released: Macaw On-Device Agent and BTL-4 Reasoning Model

maximelabonne · x · 2026-08-06

Less than two days after the release of LFM2.5-2.6B, the first fine-tuned models have arrived. Macaw 2.7B is an on-device assistant for macOS featuring 97 verified tools, taking only 1.5GB on disk with 2s response time. BTL-4 35B focuses on frontier agentic reasoning, scoring 78.4% on SWE-bench Verified and 66.1% on LiveCodeBench v6, with native 262K context. Both models are open weights.

Original post →

More from coding & agent

coding & agent channel →