@Ex0byt: Open frontier intelligence, in your hands - the MiniMax-M3 PRISM Dynamic-Quant recipe is ready! 428B parameters compres…
Summary
The MiniMax-M3 PRISM Dynamic-Quant recipe compresses a 428B parameter model from ~450GB to 119GB using per-tensor sensitivity ranking, with plans to prune further to 60-80GB for local deployment.
View Cached Full Text
Cached at: 06/15/26, 02:51 AM
Open frontier intelligence, in your hands - the MiniMax-M3 PRISM Dynamic-Quant recipe is ready! 428B parameters compressed from the ~450GB MXFP8 release down to 119GB, per-tensor sensitivity ranking that protects attention and shared-expert paths (3.4–4.5 bpw) squeezing the two largest routed-expert tensor blocks. Still too large for most local rigs, so next we prune the experts that won’t harm agentic/coding performance collapse. Target: 60–80GB.
Similar Articles
@TeksEdge: With MiniMax M3 open source now out, here is what to expect on quants and sizes, including VRAM needed: MiniMax M3 (428…
MiniMax M3, a 428B MoE model with ~23B active parameters, is now open source. It offers ultra-long context (up to 1M) and efficiency improvements, with various quantized sizes and VRAM requirements for local deployment.
@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.
@stevibe: MiniMax M2.7 is 230B params. Can you actually run it at home? I tested Unsloth's UD-IQ3_XXS (80GB) on 4 different rigs:…
A user tested MiniMax M2.7 (230B parameter model) using Unsloth's UD-IQ3_XXS quantization (80GB) across four different hardware configurations including RTX 4090, RTX 5090, RTX PRO 6000, and DGX setups, reporting token generation speeds and time-to-first-token metrics.
@TheAhmadOsman: Been playing with @PrismML's new model that turned Qwen 3.5 27B into a sub-4GB and sub-6GB weights and I am impressed C…
PrismML released a compressed version of Qwen 3.5 27B that fits in sub-4GB and sub-6GB memory, enabling impressive local AI performance.
@dealignai: MiniMax m3, made for 128gb Mac’s Thank you to @hornsby_andrew for preparing the pruning calibration dataset and doing e…
A pruned and quantized version of MiniMax-M3 (MiniMax-M3-Medium-JANG_2L) optimized to run on 128GB Macs using vMLX, featuring 32% expert pruning and JANG_2L mixed-precision quantization to fit within ~105 GB.