{"id":998,"date":"2026-05-29T07:29:49","date_gmt":"2026-05-28T23:29:49","guid":{"rendered":"https:\/\/connectword.dpdns.org\/?p=998"},"modified":"2026-05-29T07:29:49","modified_gmt":"2026-05-28T23:29:49","slug":"liquid-ai-releases-lfm2-5-8b-a1b-an-on-device-moe-model-with-8-3b-total-and-1-5b-active-parameters","status":"publish","type":"post","link":"https:\/\/connectword.dpdns.org\/?p=998","title":{"rendered":"Liquid AI Releases LFM2.5-8B-A1B: An On-Device MoE Model With 8.3B Total and 1.5B Active Parameters"},"content":{"rendered":"

Liquid AI just shipped LFM2.5-8B-A1B<\/strong>. It is an on-device Mixture-of-Experts (MoE) model built for tool calling. The model holds 8.3B total parameters but activates only 1.5B per token. That sparsity is what lets it run on consumer hardware.<\/p>\n

The release follows LFM2-8B-A1B, which Liquid AI team published earlier. LFM2.5 is a new family of hybrid models for on-device deployment. This version adds a 128K context window, reasoning, and scaled-up training. <\/p>\n

What is LFM2.5-8B-A1B <\/strong><\/h2>\n

The model uses a sparse MoE design. It activates 1.5B of 8.3B total parameters per forward pass. That keeps each generated token cheap to compute.<\/p>\n

The architecture has 24 layers. Eighteen are double-gated LIV convolution blocks; six are GQA layers. It combines MoE, GQA, and gated short convolution blocks. The context length is 131,072 tokens. The model covers nine languages, including Arabic, Chinese, and Japanese.<\/p>\n

Liquid AI team recommends a temperature of 0.2, top_k of 80, and repetition_penalty of 1.05.<\/p>\n

Unlike its predecessor, LFM2.5-8B-A1B is a reasoning-only model. It produces an explicit chain of thought before its final answer. Liquid AI team chose this because MoE models run in compute-bound settings. A smaller active parameter count makes each reasoning token inexpensive.<\/p>\n

What Changed Since LFM2-8B-A1B<\/strong><\/h2>\n

Liquid expanded the context window from 32,768 to 128,000 tokens. Pretraining scaled from 12T to 38T tokens. The vocabulary doubled from 65,536 to 128,000 tokens.<\/p>\n

The larger vocabulary tokenizes non-Latin scripts more efficiently. Liquid AI team reports the strongest compression gains in Hindi, Thai, Vietnamese, Indonesian, and Arabic. The rest of the architecture stays the same as LFM2-8B-A1B.<\/p>\n

How Liquid AI Trained It<\/strong><\/h2>\n

Liquid AI team extended the tokenizer in place rather than retraining from scratch. It continued BPE merge training from the original merges on a multilingual corpus. New embedding rows initialize as the mean of their sub-token decompositions. A brief two-stage adaptation then recovers quality.<\/p>\n

Context extension came in two phases. A 2T token midtraining phase reached 32K, focused on reasoning, math, and tool use. Raising the RoPE base \u03b8, plus a 400B token stage, reached 128K.<\/p>\n

Two reinforcement learning stages target known failure modes. A preference optimization stage reduces \u2018doom loops\u2019 in long reasoning traces. It redistributes probability mass toward plausible alternatives. A separate RL shaping reward discourages loop-inducing restart words like \u2018Wait\u2026\u2019. Another RL stage uses an avg@k-based reward to cut hallucinations. The goal is abstention on queries beyond reliable knowledge.<\/p>\n

\n
\"\"
https:\/\/www.liquid.ai\/blog\/lfm2-5-8b-a1b<\/figcaption><\/figure>\n<\/div>\n

The Benchmark Case<\/strong><\/h2>\n

LFM2.5-8B-A1B improves over its predecessor across the board. The AA-Omniscience Non-Hallucination Rate jumped from 7.46 to 63.47. IFEval rose from 79.44 to 91.84. MATH500 climbed from 74.80 to 88.76. Tau\u00b2 Telecom rose from 13.60 to 88.07.<\/p>\n

Liquid AI team compared the model against dense and MoE alternatives. On instruction following, it matches Gemma-4-26B-A4B-IT on IFEval. It does so at a fraction of the active parameter count. On Tau\u00b2 Telecom, it scores 88.07, ahead of much larger models.<\/p>\n

The avg@k reward drives a much lower hallucination rate. Accuracy stays reasonable for the model\u2019s size. On agentic benchmarks, it remains competitive with bigger models.<\/p>\n

\n\n\n\n\n\n\n\n\n
Benchmark<\/th>\nLFM2-8B-A1B<\/th>\nLFM2.5-8B-A1B<\/th>\n\u0394<\/th>\n<\/tr>\n<\/thead>\n
AA-Omniscience Non-Hallucination Rate<\/td>\n7.46<\/td>\n63.47<\/td>\n+56.01<\/td>\n<\/tr>\n
IFEval<\/td>\n79.44<\/td>\n91.84<\/td>\n+12.40<\/td>\n<\/tr>\n
MATH500<\/td>\n74.80<\/td>\n88.76<\/td>\n+13.96<\/td>\n<\/tr>\n
Tau\u00b2 Telecom<\/td>\n13.60<\/td>\n88.07<\/td>\n+74.47<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/figure>\n

Running It: CPU, GPU, and Tooling<\/strong><\/h2>\n

The model ships with day-one support across the inference ecosystem. Frameworks include llama.cpp, MLX, vLLM, and SGLang. ONNX and Liquid\u2019s LEAP edge platform are also supported.<\/p>\n

On CPU, it decodes 253 tokens\/s on an M5 Max. It reaches 146 tokens\/s on a Ryzen AI Max+ 395. It stays under 6 GB of memory throughout. On a phone, it holds about 30 tokens\/s.<\/p>\n

On a single NVIDIA H100 SXM5, output throughput hits 18.5K tokens per second. That is over 1.6B tokens per day at high concurrency.<\/p>\n

For tool use, LFM2.5 writes Pythonic function calls by default. They appear between the <|tool_call_start|><\/code> and <|tool_call_end|><\/code> special tokens. You can override this to JSON in the system prompt.<\/p>\n

Strengths <\/strong>and What to Watch<\/strong><\/h2>\n

Strengths:<\/strong><\/h4>\n