MiniMax has released a new artificial intelligence model called M3.1 Flash, according to a report by MarkTechPost. The model features a 428-billion-parameter Mixture of Experts architecture, utilizing roughly 23 billion active parameters. It also supports a context window of one million tokens. API pricing for the system is set at $0.15 per million input tokens and $0.60 per million output tokens, with model weights scheduled for release on October 23.
Mixture of Experts Architecture and Pricing
The Mixture of Experts approach allows large language models to activate only a subset of their total parameters for any given task, which typically reduces computational costs during inference. This efficiency likely enables the aggressive pricing structure announced for the M3.1 Flash API. High-capacity context windows paired with low per-token costs represent a growing trend among AI developers aiming to capture enterprise market share for complex, agentic workflows that require processing vast amounts of text simultaneously.
Unresolved Details and What to Watch
While the announcement outlines the parameter count, context length, and pricing, the provided details leave several operational questions unanswered. The announcement did not specify the exact benchmark performance scores of M3.1 Flash compared to competing models from OpenAI, Anthropic, or open-weight alternatives. Additionally, the announcement did not specify the hardware infrastructure required to host the model locally once the weights are released, nor did it detail the specific agentic capabilities built into the system. Observers will need to monitor the October 23 weight release to evaluate the model’s practical utility.
As the release date approaches, developer benchmarks and independent evaluations will clarify how the 428B model performs in real-world agentic scenarios.
Lead image: panumas nikhomkhai / Pexels
