@maximelabonne: You can't stop us from going smaller.
Summary
A tweet discusses optimizing a local LFM2.5-2.6B AI model by switching from F16 to QAD Q4_0 quantization, resulting in reduced size, faster speeds, and lower latency while maintaining high performance.
View Cached Full Text
Cached at: 08/20/26, 02:59 PM
You can’t stop us from going smaller.
Noctus (@noctus91): So you’re telling me I can swap my local LFM2.5-2.6B from F16 to QAD Q4_0 and go from:
5.4 GB → 1.6 GB 21 → 64 tok/s 3.0s → 1.2s tool-call latency
while keeping ~97% of BF16 performance?
@LiquidAI what did you just do 😭
Similar Articles
@noctus91: I recently switched from Qwen 3.5 9B to LFM2.5-8B-A1B by @liquidai, and it's quickly become my default local model in H…
A user shares their positive experience switching from Qwen 3.5 9B to Liquid AI's new LFM2.5-8B-A1B model, praising its speed and reliability for agentic tasks while noting coding remains a weakness. The model is an 8B MoE with 1.5B active parameters and 128K context, optimized for devices and server-side use.
@no_stp_on_snek: Config-I quant of MiniMax-M3 is up on MLX. 2-bit experts, 4-bit attention, 8-bit boundaries + embeddings, f16 router. ~…
Announces the release of a Config-I quantization of MiniMax-M3 on MLX, using 2-bit experts and 4-bit attention to reduce the 427B MoE model from 869GB to ~167GB, though the quant is untested and requires a patch for mlx_lm.
@liquidai: Introducing LFM2.5-230M: our smallest model yet, built to run fast anywhere (CPUs, NPUs, and GPUs) to enable agentic ta…
Liquid AI releases LFM2.5-230M, a small 230M parameter model optimized for fast inference on CPUs, NPUs, and GPUs, targeting agentic tasks on devices like phones and robots.
@maximelabonne: Wow, this gives me flashbacks of early model merging. Complete insanity, I love it!
A new sub-6B sparse activation AI model built with a fusion architecture combines weights from LFM2.5-2.6B and Qwen3.6-35B-A3B, achieving near-Qwen3.6-35B performance at a fraction of the size.
Stop asking what model to run. There are literally only two.
A tech enthusiast argues that only two local AI models (Qwen 3.6 35b a3b and Qwen 3.6 27b) are worth running, dismissing smaller models and recommending heavy quantization of larger models.