Liquid AI releases LFM3 open-weight multimodal model family on hybrid backbone
Liquid AI announces LFM3, a new generation of Liquid Foundation Models spanning text, vision, audio, omni, and decision models on a shared hybrid backbone. The first release covers 2B-class models, pre-trained on 22T tokens. The base and post-trained models are open-weight and available on Hugging Face, the Playground, and the Liquid AI API. The architecture was redesigned via hardware-in-the-loop search across 14 device and backend combinations. Gated short convolutions remain over half the layers, with sliding-window attention replacing some for better long-context speed.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · automated brief
- Discovery linkX
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
Liquid AI announces LFM3, a new generation of Liquid Foundation Models spanning text, vision, audio, omni, and decision models on a shared hybrid backbone. The first release covers 2B-class models, pre-trained on 22T tokens. The base and post-trained models are open-weight and available on Hugging Face, the Playground, and the Liquid AI API. The architecture was redesigned via hardware-in-the-loop search across 14 device and backend combinations. Gated short convolutions remain over half the layers, with sliding-window attention replacing some for better long-context speed.
Key findings from the stored brief
- LFM3 is released as six 2B-class models across text, vision, audio, omni, and decision modalities on one shared backbone.
- Pre-training used 22T tokens, and the base and post-trained models are open-weight on Hugging Face.
- Architecture search spanned 14 device and backend combinations, including phones, laptops, Raspberry Pi 5, cloud CPUs, GPUs, and a Qualcomm NPU.
- Per the source, LFM3-1.7B matches LFM2-1.2B per parameter on short prompts and is about 25% faster at 8K context.
Source claims not independently checked; performance figures come from the vendor's own benchmarks.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
These are Grok Bot discovery notes. AINewsIQ prepared a brief from available source material. The bot's rationale, linked posts, authors, and stances have not been independently verified.
Liquid AI's Oct 27 post introduces LFM3, six 2B-class models (text, vision, audio, omni, decision) trained from LFM3-1.7B-Base on 22T tokens with a new convolution/sliding-window/global-attention hybrid. It says the text model fits in about 1 GB at 4-bit, has 256K context, and beats LFM2.5-2.6B. Weights are on Hugging Face under Liquid's LFM 1.0 license. Benchmarks are vendor-reported.
Source claims not independently checked; performance figures come from the vendor's own benchmarks.
Engagement counts, when shown, are observed snapshots. They do not establish accuracy or corroboration.
- hn post ↗ · 88 points, 32 comments · observed 10/27/2026, 10:09:31 PM
- x post ↗ · 3,263 likes, 996 reposts, 272 replies, 303 quotes · observed 10/27/2026, 10:09:03 PM
- Related post ↗
- Related post ↗
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
Open-weight, small multimodal models that run on-device and can be fine-tuned locally lower barriers for specialized, privacy-sensitive, or edge deployments. Efficiency gains at long context matter for agentic workloads.
Limits and open questions
Source claims not independently checked; performance figures come from the vendor's own benchmarks.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.