MiniMax M3
MiniMax's open-weight multimodal MoE — 428B total parameters with ~23B active per token, a 1M-token context, and native image and video input. Built for sustained agentic coding rather than single-shot answers.
What MiniMax M3 is best for
Sustained multi-step agentic coding, visual debugging, long-context codebase analysis, multimodal tasks.
Strengths
428B total parameters (MoE, ~23B active per token)
Native multimodal input — text, image, and video in the same call
1M-token context (512K guaranteed, 1M ceiling)
66.0% on Terminal-Bench 2.1 — strongest sustained-agent score in its price class
~78% on SWE-bench Verified
MSA sparse attention — 9-15x faster than the previous M2 generation
Open weights on Hugging Face, so it can be self-hosted
Served by five independent providers on Multos, with automatic failover
When to use MiniMax M3
Choose MiniMax M3 when a task runs long and needs to stay coherent — multi-hour agentic coding sessions, reproducing a paper, or working through a large codebase across many sequential tool calls. It is also the right pick when you need to reason about something visual: pass a UI screenshot or a screen recording and ask what the user was doing. Because the weights are open and five providers serve it, it is a reasonable default when you want frontier-adjacent capability without depending on a single vendor.
Limitations to know
Output is text only — it accepts images and video but does not generate them. At $0.30/$1.20 per million tokens it costs more than DeepSeek V4 Flash ($0.14/$0.28) and GPT-5.6 Luna ($0.20/$1.20), so for high-volume classification, extraction, or routing work a cheaper model is the better fit. Its blended cost at production cache rates is roughly $0.14/M, the highest of the three efficient-tier options.
Where MiniMax M3 sits in the lineup
Multos AI groups models into four access levels. Higher levels unlock stronger models; every level draws from the same per-message model switcher.
Free & Lite
7
models
Starter+
12
models
Pro+
9
models
Elite
7
models
Questions about MiniMax M3
What is MiniMax M3?
MiniMax M3 is a MiniMax model available on Multos AI. MiniMax's open-weight multimodal MoE — 428B total parameters with ~23B active per token, a 1M-token context, and native image and video input. Built for sustained agentic coding rather than single-shot answers.
What is MiniMax M3's context window?
MiniMax M3 supports a 1M tokens context window on Multos AI, which determines how much code, documentation, and conversation history it can hold in a single request.
What plan do I need for MiniMax M3?
MiniMax M3 requires the Starter+ plan or higher. Free users can access lite-tier models immediately and upgrade to reach MiniMax M3.
What is MiniMax M3 best for?
MiniMax M3 is best for sustained multi-step agentic coding, visual debugging, long-context codebase analysis, multimodal tasks.
Try MiniMax M3 on Multos AI
Every model on one subscription. Switch per message. Start free.
Get Started Free