MiniMax M3 — Specifications
Compiled by Mustafa Ihsan from the vendor’s published documentation · Last updated
| Developer | MiniMax |
|---|---|
| Type | LLM (MoE, coding/agentic, multimodal) |
| Modality | Text, Image, Video → Text |
| Parameters | ~428B total / ~23B active (MoE) |
| Context window | 1M |
| License | MiniMax Community License (open weights) |
| Open weights | Yes |
| Released | 2026-06 |
| Input price | $0.30 /1M |
| Output price | $1.20 /1M |
| API providers | MiniMax API, OpenRouter |
Benchmarks
| MiniMax-reported | SWE-Bench Pro 59.0%; Terminal-Bench 2.1 66.0%; MCP Atlas 74.2%; BrowseComp 83.5. Prices shown are for requests up to 512K input tokens; $0.60 / $2.40 above that. |
|---|
What is MiniMax M3?
MiniMax M3 is MiniMax’s large language model for coding and agentic work, released on 1 June 2026. According to MiniMax, it combines frontier-level coding and agent skills, a context window of up to 1M tokens built on its new MiniMax Sparse Attention (MSA), and native multimodality with image and video input. MiniMax says that at a 1M-token context, M3’s per-token compute is 1/20 that of the previous generation. It is a mixture-of-experts with about 428B total and 23B active parameters, and MiniMax reports 59.0% on SWE-Bench Pro and 66.0% on Terminal-Bench 2.1.
MiniMax’s API price list is tiered by prompt length. For requests up to 512K input tokens, M3 costs $0.30 per million input tokens, $1.20 per million output tokens and $0.06 per million cached reads; above 512K it costs $0.60 / $2.40 ($0.12 cached). MiniMax shows these as a permanent 50% discount on list prices, and its Priority tier, for faster and more reliable admission, costs 1.5× the standard rate. MiniMax describes the API context as up to 1M tokens with a guaranteed minimum of 512K.
The weights are on Hugging Face under the MiniMax Community License: free for non-commercial use, while commercial users must display “Built with MiniMax M3” and notify MiniMax, or get written permission above $20 million in yearly revenue. On 19 October 2026 MiniMax released the cheaper M3.1-Flash ($0.15 / $0.60), now its latest M-series model; M3 remains available. Up to 512K input, M3 matches the peak price of DeepSeek V4.1-Flash. MiniMax’s video model is MiniMax H3, and the AI API cost calculator estimates monthly costs.
MiniMax M3 pricing: API cost per 1M tokens
| Input (per 1M tokens) | $0.300 |
|---|---|
| Output (per 1M tokens) | $1.20 |
| Output/input ratio | 4× |
| Blended (4:1 in:out) | $0.480 per 1M tokens |
What MiniMax M3 costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Workload | Tokens / month | Cost / month |
|---|---|---|
| Side project | 1M in / 0.25M out | $0.60 |
| Small team | 20M in / 5M out | $12 |
| Production | 200M in / 50M out | $120 |
Run your own numbers in the AI API cost calculator.
Cheaper alternatives to MiniMax M3
| Model | Blended $/1M | You save |
|---|---|---|
| Llama 3.1 8B open | $0.0240 | 95% cheaper |
| Mistral NeMo 12B open | $0.0292 | 94% cheaper |
| Gemma 3 4B open | $0.0600 | 88% cheaper |
Frequently asked questions
How much does MiniMax M3 cost per 1M tokens?
MiniMax M3 costs $0.300 per 1M input tokens and $1.20 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $0.480 per 1M tokens.
How much does MiniMax M3 cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $12 on MiniMax M3. A side project (1M in / 0.25M out) costs roughly $0.60.
What is a cheaper alternative to MiniMax M3?
Llama 3.1 8B is the strongest cheaper option in our database at $0.0240 per 1M blended — about 95% less than MiniMax M3. It is also open-weight, so self-hosting is an option.
Can I run MiniMax M3 locally?
Yes. MiniMax M3 is open-weight and needs about a large amount of VRAM at 4-bit quantisation.
Why does MiniMax M3 charge more for output than input?
Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost the provider more to serve. MiniMax M3 charges 4× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
Prices are the published list rates for the model's primary API and are reviewed as providers change them. Volume, batch and cached-input discounts are not included. Compare every model side by side in the AI models database or the LLM leaderboard.