MiniMax
M3 flagship with 1M context + MSA attention + Token Plan in 3 tiers from ¥49/¥119/¥469
Why we recommend MiniMax
Why we recommend MiniMax
MiniMax (Shanghai Xiyu Technology / 上海稀宇科技有限公司, brand "MiniMax") is one of China's leading AGI labs — founded in early 2022 — building self-developed multimodal foundation models spanning language / video / voice / image / music. The latest flagship (released 2026-05) is M3: a brand-new MSA attention architecture, native support for 1M ultra-long context + multimodal input (image / video), and a state-of-the-art model for Coding / Agentic workloads.
One MiniMax, multiple domains. MiniMax splits across multiple surfaces:
platform.minimaxi.comis the open platform (API console + Token Plan subscription) — the "Subscribe on minimax" buttons jump here.www.minimaxi.comis the brand portal (company info + model lineup + latest blog).code.minimaxi.comis the MiniMax Code desktop (Coding Agent tuned for MiniMax models).hub.minimaxi.comis the MiniMax Hub (creative playground).www.minimaxi.com/audiois the MiniMax Audio (TTS studio).www.xingyeai.comis Xingye (companion social AI product).All run on the same MiniMax account — sign up once and you can use all of them.
M3 flagship + Token Plan 3 tiers + Credit Packs 3 tiers:
| Tier | Monthly | Tokens / month | Concurrent agents | Video quota |
|---|---|---|---|---|
| Plus | ¥49 / mo | ~600M+ | 3-4 agents | × |
| Max | ¥119 / mo | ~1.8B+ | 4-5 agents | 3 / day |
| Ultra | ¥469 / mo | ~7.1B+ | 6-7 agents | 5 / day |
| Credit Pack Starter | ¥30 / 4,285 credits | — | — | 365-day TTL |
| Credit Pack Mid | ¥150 / 21,430 credits | — | — | 365-day TTL |
| Credit Pack Pro | ¥500 / 71,435 credits | — | — | 365-day TTL |
Highlights:
- M3 flagship, 1M context, brand-new MSA attention: native multimodal (image / video input); pay-as-you-go with ≤512k context at permanent 50% off — ¥2.10 / M input tokens, ¥8.40 / M output tokens.
- H3 video model (released 2026-07): 2K resolution + 4-15s duration, supporting T2V / I2V / first-last frame / multimodal reference; Token Plan Max and above include daily video-generation quota.
- M2.7 / M2.7-highspeed dual options: self-iterating reasoning + high-speed variant. M2.7 input ¥2.10 / M tokens, M2.7-highspeed input ¥4.20 / M tokens.
- Full multimodal stack: Speech-2.8-HD (emotion rendering) / Turbo (extreme speed) + Music-3.0 (more natural vocal synthesis) + image-01 (delicate rendering) + video generation (Hailuo series).
- Complete pay-as-you-go coverage: credit packs at 1,000 credits = ¥7 with 365-day validity; when subscription quota is exhausted, credits automatically top up overage.
- Subscription + Credit Pack dual channels: when the Token Plan quota runs out, you can keep going on demand via credit packs at the standard 1,000 credits = ¥7 rate — useful for occasional traffic spikes + long-term cost control.
In practice, subscription entry is at platform.minimaxi.com/subscribe/token-plan — new users get a 0-yuan trial across Plus / Max / Ultra. For production workloads, switch to pay-as-you-go at pricing-paygo.
If your project needs 1M long-context Coding model + native multimodal understanding + video generation (2K) + full multimodal stack (voice / music / image) + value subscription, MiniMax's Token Plan is a strong 2026 pick.