Liquid AI Releases LFM3 Family of Six Open-Weight Models Built for Single-GPU Use

Liquid AI launched the LFM3 family, six open-weight multimodal models of roughly 2 billion parameters each, on October 27, 2026. The models are built to run and be fine-tuned on a single GPU. Crypto Briefing reported the launch, which the company framed around local deployment rather than data-center scale.

Inside the Six-Model LFM3 Lineup

All six models share a common foundation called LFM3-1.7B-Base, with each variant tuned for a different kind of input. The lineup spans text, vision-language, audio, decision-making and an "omni" model that handles text, images, video and audio together. The text model, LFM3-1.7B, supports a context window of 256K tokens.

The vision-language model, LFM3-VL-2B, is where the company makes its boldest claim. According to the research findings, it has shown performance matching models more than six times its size on vision tasks. Because the models are open-weight, developers can download and adapt them, and Liquid AI highlights domain-specific uses such as robotics and defect detection.

The technical centerpiece is a tri-hybrid architecture combining gated short convolutions, sliding-window attention and global attention. The architecture was pre-trained on 22 trillion tokens, and the research findings describe a high performance-per-parameter ratio compared with larger models. Liquid AI was founded in 2023 and has raised approximately $287.5 million since then, including a $250 million Series A round in December 2024.

Liquid AI's Funding and Partnership Record

Liquid AI's akta.pro company records list three funding rounds, including a $250 million Series A in December 2024 led by AMD Ventures and a $37.5 million seed in December 2023 led by OSS Capital L.P. and The Pags Group. The same records show a $2.4 billion unicorn valuation reached after the Series A and $13.2 million in annualized recurring revenue reported for 2025.

The records also list a multi-year strategic partnership with Mercedes-Benz, entered in April 2026, to embed Liquid AI's on-device foundation models into MBUX vehicles, with production deployment targeted for the second half of 2026 on third- and fourth-generation MBUX vehicles back to the 2024 model year.

Earlier Launches From Liquid AI

Liquid AI announced RIPL, a real-time inference engine for its LFM2.5 models, on October 22, 2026, claiming sub-50 ms end-to-end latency, MarkTechPost reported. Benchmarks cited 5.6 to 28.4x lower p95 latency than Qwen3.5 on vLLM, with Shopify using it for sub-20 ms search workloads.

On October 15, 2026, Liquid AI launched its d1-ultra decision model, which returns calibrated probabilities from text or images without generating output tokens, The future of trading reported. On August 13, 2026, MarkTechPost reported that Liquid AI released LFM2.5-VL-3B, a 3.1-billion-parameter vision-language model designed for on-device deployment with approximately 3 GB of memory usage.

Mercedes-Benz Deployment Target

The Mercedes-Benz partnership recorded in April 2026 targets first production deployment in the second half of 2026, with availability on third- and fourth-generation MBUX vehicles back to the 2024 model year.