MiniMax M3 is a MiniMax model with 1M context, $0.30 in / $1.20 out per million tokens, last verified October 25, 2026. Open weights: yes. Public-board scores are linked to the publisher; vendor-only figures stay in the watch-out, not on the board.
MiniMax M3 is an open-weight (community license) 428B MoE with native multimodal and 1M context.
Recorded profile verification October 25, 2026·Individual provider fields may change
Pricing tiers: MiniMax PAYG USD/MTok Standard input/output $0.30/$1.20 at <=512K input, cache reads $0.06; above 512K input $0.60/$2.40, cache $0.12. Priority short input/output/cache $0.45/$1.80/$0.09, long $0.90/$3.60/$0.18; default Standard, service_tier=priority selects priority admission. The pricing page labels these current rates permanent 50% off; no future rate is established. PAYG balance and subscription Credits/M Plan are separate. M3.1 Flash has a separate tariff and does not accept service_tier.
Publisher evidence
Published benchmark results
Each result keeps its source date visible. Dates refer to source publication or capture unless the source explicitly states when it measured the result. A missing benchmark is not treated as a zero.
MiniMax vendor report: Terminus2, 8CPU/16GB sandbox, 2h, 128K generation budget; budget is not an API cap · View source
Best for
Open-weight deployments
Multimodal
Long-context
Watch out
Commercial checkpoint use requires Built with MiniMax M3 attribution and a one-time notice at or below $20M yearly revenue from the relevant products/services; above that threshold prior written authorization is required. Keep copyright/permission notices and follow prohibited-use terms. The checkpoint documents enabled/adaptive/disabled thinking. Hosted OpenAI compatibility defaults thinking on; Anthropic compatibility defaults it off and enables it with adaptive. Preview effort levels do not apply to M3. A hard output cap remains unverified. Current compatibility guides recommend max_completion_tokens for new OpenAI integrations; reasoning counts toward completion limits. Image/video input is supported, audio input is not. Unsupported OpenAI sampling parameters can be ignored by the API even though native inference recommends temperature=1/top_p=0.95.
Source receipts
Catalog figures for this model were checked against the following sources.
Answered from the verified figures on this page rather than general guidance.
How much does MiniMax M3 cost per million tokens?
MiniMax M3 lists base rates of $0.30 per million input tokens and $1.20 per million output tokens. Estimate both input and output token volume and any additional charges for your workload. MiniMax PAYG USD/MTok Standard input/output $0.30/$1.20 at <=512K input, cache reads $0.06; above 512K input $0.60/$2.40, cache $0.12. Priority short input/output/cache $0.45/$1.80/$0.09, long $0.90/$3.60/$0.18; default Standard, service_tier=priority selects priority admission. The pricing page labels these current rates permanent 50% off; no future rate is established. PAYG balance and subscription Credits/M Plan are separate. M3.1 Flash has a separate tariff and does not accept service_tier.
What is MiniMax M3's context window?
MiniMax M3 accepts about 1M tokens of context. That only matters if you routinely send very long documents, large codebases, or multi-turn histories that approach that limit.
What is MiniMax M3 best for?
MiniMax M3 is a frontier tier from MiniMax. It suits open-weight deployments, multimodal, long-context. Commercial checkpoint use requires Built with MiniMax M3 attribution and a one-time notice at or below $20M yearly revenue from the relevant products/services; above that threshold prior written authorization is required. Keep copyright/permission notices and follow prohibited-use terms. The checkpoint documents enabled/adaptive/disabled thinking. Hosted OpenAI compatibility defaults thinking on; Anthropic compatibility defaults it off and enables it with adaptive. Preview effort levels do not apply to M3. A hard output cap remains unverified. Current compatibility guides recommend max_completion_tokens for new OpenAI integrations; reasoning counts toward completion limits. Image/video input is supported, audio input is not. Unsupported OpenAI sampling parameters can be ignored by the API even though native inference recommends temperature=1/top_p=0.95.
Can I self-host MiniMax M3?
MiniMax M3 publishes open weights, but self-hosting depends on the licence, hardware footprint, quantisation quality, and serving stack. A hosted API is often cheaper until you have measured throughput and concurrency on your own hardware.
