Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Official7
liquid.ai2026-05-28officialHN front pageabout the LFM2 familyHN 245 pts · 96 comments1,792 words
Liquid AI released LFM2.5-8B-A1B, an edge model designed for fast tool calling on consumer hardware. It features a 128,000 token context window and scaled pretraining to 38T tokens, expanding its vocabulary from 65,536 to 128,000 to improve non-Latin tokenization. The model decodes at 253 tokens/s on an M5 Max and 146 on a Ryzen AI Max+ 395 while staying under 6 GB, reaching 18.5K output tokens per second on a single H100.written, reviewed, cited
- The model doubles its vocabulary to 128K by extending the existing tokenizer in place rather than retraining the model from scratch.
- A targeted preference optimization stage identifies tokens that tend to trigger looping behavior in specific contexts to reduce doom loops.
- The model achieves 18.5K output tokens per second at high concurrency on a single NVIDIA H100 SXM5 GPU.
liquid.ai2026-08-12officialabout the LFM2 familyHN 11 pts · 3 comments2,057 words
Liquid AI releases LFM2.5-VL-3B, a non-reasoning vision-language model that integrates a SigLIP2 400M NaFlex encoder and doubles its vocabulary to 128K. It pre-trains on ~34T tokens and decodes 228 tokens/s on an Apple M5 Max. The model achieves about 11K tokens per second output throughput on a single NVIDIA H100.written, reviewed, cited
- The model doubles its vocabulary to 128K by extending the existing tokenizer in place to support non-Latin scripts.
- LFM2.5-VL-3B runs on a phone, decoding 20 tokens/s on a Galaxy S26 Ultra for private on-device execution.
- Evaluation uses vLLM 0.26.0 and non-reasoning mode where models are prompted to directly answer without reasoning.
liquid.ai2026-07-01officialabout the LFM2 familyHN 20 pts · 1 comments933 words
Liquid AI releases LFM2.5-230M, a 230M-parameter model pre-trained for 19T tokens that runs on-device on hardware like the Unitree G1 humanoid robot using NVIDIA's SONIC framework. It achieves 213 tok/s decode speed on a Galaxy S25 Ultra and 42 tok/s on a Raspberry Pi 5 while competing with larger models in knowledge and instruction following benchmarks. Although capable of tool use and data extraction, the model is not recommended for reasoning-heavy workloads such as advanced math or code generation.written, reviewed, cited
- The post-training recipe includes multi-domain reinforcement learning to balance out-of-the-box capabilities with downstream adaptability.
- Engineers must tune the flash-attention flag per device, enabling it on Raspberry Pi 5 but disabling it on Snapdragon Gen4.
huggingface.co2026-08-06authorityabout the LFM2 familyHN 1 pts · 0 comments1,103 words
Liquid AI releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, two new models designed for fast long-context inference on CPU. These encoders support an 8,192-token context and are about 3.7× faster than ModernBERT-base at long context lengths. The models are trained with a masked language objective that masks 30% of tokens during training to achieve strong performance on GLUE, SuperGLUE, and multilingual tasks.written, reviewed, cited
- The 350M variant is the choice when accuracy matters most.
- The 230M variant is the choice for tighter hardware or higher throughput.
- A base encoder provides general-purpose representations rather than direct task outputs.
liquid.ai2026-08-04officialabout the LFM2 familyHN 3 pts · 0 comments1,566 words
Liquid AI releases LFM2.5-2.6B, a 2.6B parameter agentic model trained on ~34T tokens and available on Hugging Face. It features a 128K vocabulary and context-extension phase, decoding 220 tokens/s on an M5 Max and 113 tokens/s on a Ryzen AI Max+ 395 while staying under 2.5 GB. The model reaches 15K output tokens per second on an H100 GPU.written, reviewed, cited
- The model doubles its vocabulary to 128K by extending the existing tokenizer in place rather than retraining the model from scratch.
- Training uses a dedicated 128K context-extension phase so the model can handle the long inputs that agentic workloads require.
- The training pipeline separates model optimization, inference, and environment execution into distinct components using FSDP, SGLang, and verl.
liquid.ai2026-04-01officialabout the LFM2 familyHN 3 pts · 1 comments1,629 words
liquid.ai2026-04-27officialabout the LFM2 familyHN 2 pts · 0 comments956 words
Analysis2
2tinteractive.com2026-09-24about the LFM2 familyHN 1 pts · 1 comments888 words
LiquidAI released LFM2.5-VL-DSpark on September 24, 2026, which accelerates vision-language model training from 20 days to 3 days. The architecture enhances computational efficiency by 20 percent and includes new training data sets curated by experts. It features integration APIs and GDPR compliance details available on Hugging Face.written, reviewed, cited
- The model includes a new neural architecture designed to minimize latency for real-time video analysis.
- The architecture is more energy-efficient, reducing the computational resources and costs associated with training.
blog.mozilla.ai2026-03-19about the LFM2 familyHN 7 pts · 0 comments629 words
