Advertisement
Verified Partner
Enterprise Cloud Infrastructure, GPU Clusters & AI APIs
Deploy high-throughput inference and quant trading pipelines with ultra-low latency.
Explore Platform β†’

MiniMax Launches MiniMax-M3 and H3: MXFP8 8-bit Microscaling and Music3 Multimodal Generation

By Dr. Sarah Chen β€’ Published on 2026-09-07 β€’ 4 min read β€’ Source: MiniMax AI
MiniMax releases MiniMax-M3 in hardware-accelerated MXFP8 format, delivering lightning-fast inference for conversational agents and high-fidelity audio synthesis.

MiniMax has made a series of major releases on Hugging Face, including **MiniMax-M3** (with native MXFP8 format) and **MiniMax-Music3**.

Microscaling FP8 Precision By adopting the OCP Microscaling (MX) FP8 specification, MiniMax-M3 reduces memory bandwidth consumption by 50% compared to standard BF16, while preserving 99.8% of mathematical accuracy. This allows sub-50ms conversational latencies in real-time voice agents.

Advertisement
Verified Partner
Quantitative Trading Systems & 30 AI Business Blueprints
Build predictable monthly recurring revenue with retainers & automated bots.
View Blueprints β†’

Source & Fact Check

This technical dispatch was verified against primary documentation released by MiniMax AI.

Read Original Announcement on MiniMax AI β†’