Liquid AI ships LFM2.5-24B-A2B: a 2B-active local agent model hitting 21K tok/s on one B300
JosephJacks_ · x · 2026-10-14
Liquid AI released LFM2.5-24B-A2B, the largest member of its LFM2.5 family, built to run multi-step agent tasks locally on laptops and PCs. Open weights are on Hugging Face.
- Specs: 24B total parameters with 2B active per token, 131K context, 128K vocabulary, pretrained on 33T tokens; a reasoning model that's easy to fine-tune
- Benchmarks: competitive with Qwen3.6-35B-A3B (50% more total and active params) on agentic/tool-use tasks — 48.09 vs 38.19 on LiveClaw, 27.69 vs 8.72 on AutomationBench
- Serving: SGLang offers Day-0 support; up to 21K output tok/s on a single NVIDIA B300, servable with one sglang serve command
Related event: LiquidAI Releases LFM2.5-24B-A2B, an On-Device MoE Reasoning Model(13 posts)→
More from Models
- Nathan Lambert slams Meta's Muse Open License as 'trash' — natolambert · 2026-10-16
- Google launches Gemini 4 Argon to Ultra and Pro subscribers, touting physics and 3D skills — ShikharMurty · 2026-10-16
- Google gives X users early access to test Argon, not just companies — lyraxana · 2026-10-16
- Grok Voice 3.0 Tops Design Arena Speech Leaderboard, Up 7.4 Points — XFreeze · 2026-10-16
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16
- Meta open-sources Muse Spark: top US open model, scores just below DeepSeek — bindureddy · 2026-10-16