Today for AI
HOT RADAR
llmHEAT 27.3°

AI CURATED HOT STORY · 10/28/2026

Liquid AI Releases LFM3, Outperforming Predecessors with Smaller Models

4 reports archived4 independent sourcesupdated 10/28/2026, 17:49:57
Synthesis & Latest Updates
4 Sources Cross-Validated

Liquid AI has officially launched the LFM3 open-weight model family, marking a key breakthrough in balancing efficiency and capability for edge-side AI. The series includes six instances covering text, vision, audio, and multimodal tasks. Its core highlight is the 1.7B parameter text model, which significantly outperforms its 2.6B predecessor and competitors in various benchmarks while supporting a 256K context window.

LFM3 utilizes an innovative "tri-mix" architecture combining gated short convolutions, sliding window attention, and global attention, optimized via hardware-in-the-loop search. Cross-validation shows superior performance in tool calling (BFCL V4) and agentic tasks (AutomationBench, WebArena), surpassing Qwen3.5-9B in specific metrics. However, it lags in long-context reasoning (RULER) and complex knowledge QA (GPQA), reflecting a strategic trade-off for scenario-specific optimization.

Released under the LFM Open License v1.0, allowing free commercial use for companies with revenue under $10 million, the models are fully open-weight and support local fine-tuning. LFM3 offers developers new options for building efficient specialized agents, potentially accelerating the shift of AI from cloud to edge devices and reshaping the on-device multimodal ecosystem.

LATEST/LFM3 weights are now fully open-source on Hugging Face, with the community actively testing local fine-tuning capabilities. Despite limitations in general knowledge QA, its high efficiency in agentic and tool-calling scenarios has sparked widespread developer interest in building lightweight specialized agents.

Hourly heat curve

42 fully observed hours
Now
27.3
Peak
50.106:00
24h change
-43%

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 4 reports · Latest first
  2. Marktechpost (@Marktechpost)T2·85 pts

    Liquid AI Releases LFM3: Six Open-Weight Multimodal Models on Shared Hybrid Backbone

    • LFM3 utilizes a tri-hybrid architecture (13 gated short conv layers + 5 sliding-window attention + 6 global attention), optimized via hardware-in-the-loop search for edge devices.
    • The 1.7B variant significantly outperforms Qwen3.5-9B on agentic benchmarks like BFCL V4 (71.39 vs 62.83) and WebArena (38.42 vs 28.94), making it suitable for tool-calling scenarios.
    • Clear trade-offs exist: It lags behind Qwen3.5-9B in complex knowledge reasoning (GPQA Diamond) and long-context performance (RULER 128K).
    • Weights are released under the LFM Open License v1.0, allowing free commercial use for companies with annual revenue below $10 million.
  3. Rohan Paul (@rohanpaul_ai)T2·85 pts

    Liquid AI Launches Open-Weight LFM3 Family with Tri-Hybrid Architecture

    • LFM3 uses a tri-hybrid architecture (short convolutions + sliding-window attention + global attention) to boost small-model efficiency.
    • The 1.7B text model scores 41.29 on AutomationBench, significantly outperforming Qwen3.5-9B's 15.10.
    • The family includes six model instances covering text, vision-language, audio, and omni-modalities, optimized for on-device execution.
  4. Hacker News AIT2·78 pts

    Liquid AI Releases Open-Weight Multimodal LFM3 Models for Local Fine-Tuning

    • LFM3 is the first 2B-parameter multimodal family unifying text, vision, audio, and decision models onto a single shared backbone.
    • Built on a new hybrid architecture with 22T tokens pre-training, it offers greater expressiveness than LFM2 while maintaining high efficiency.
    • Fully open-weight models support running on any device and local fine-tuning, ideal for building specialized intelligent agents.