MiniMax API Pricing
The Shanghai lab behind the agentic MiniMax M2 models and the Hailuo video/audio generators. Founded in 2021 by Yan Junjie (a former SenseTime research lead), MiniMax pairs cheap long-context text models with consumer media apps — and listed in Hong Kong in January 2026.
MiniMax M3 LAUNCHED JUN 2026
MiniMax M3 standard ≤512K row: $0.30/M input, $1.20/M output, $0.06/M prompt-cache read.
MiniMax M2.7 LAUNCHED MAR 2026
Beginning the journey of recursive self-improvement.
MiniMax M2.7 Highspeed
Same M2.7 family with higher throughput.
MiniMax M2.5
Peak Performance.
MiniMax M2.5 Highspeed
Same M2.5 capability with faster output.
MiniMax M2.1
Powerful multi-language programming capabilities.
MiniMax M2.1 Highspeed
Faster and more agile M2.1.
MiniMax M2
Agentic capabilities and advanced reasoning.
MiniMax M2-her
Dialogue-specialised model for role-play and conversation history.
MiniMax M1
Legacy MiniMax M1 with 1M context.
MiniMax-01
Legacy MiniMax-01 with 1M context and MiniMax-VL-01 image understanding.
| Model | Input /M | Output /M | Cached | Context | Max output | Vision | Tools | Tier |
|---|---|---|---|---|---|---|---|---|
| MiniMax M3 FLAGSHIP | $0.3 | $1.20 | $0.06−80% | 1M | — | ✓ | ✓ | Frontier |
| MiniMax M2.7 | $0.3 | $1.20 | $0.06−80% | 205K | — | ✗ | ✓ | Mid |
| MiniMax M2.7 Highspeed | $0.6 | $2.40 | $0.06−90% | 205K | — | ✗ | ✓ | Mid |
| MiniMax M2.5 | $0.3 | $1.20 | $0.03−90% | 205K | — | ✗ | ✓ | Mid |
| MiniMax M2.5 Highspeed | $0.6 | $2.40 | $0.03−95% | 205K | — | ✗ | ✓ | Mid |
| MiniMax M2.1 | $0.3 | $1.20 | $0.03−90% | 205K | — | ✗ | ✓ | Active |
| MiniMax M2.1 Highspeed | $0.6 | $2.40 | $0.03−95% | 205K | — | ✗ | ✓ | Active |
| MiniMax M2 | $0.3 | $1.20 | $0.03−90% | 205K | — | ✗ | ✓ | Active |
| MiniMax M2-her | $0.3 | $1.20 | — | 64K | — | ✗ | ✓ | Mid |
| MiniMax M1 LEGACY | $0.4 | $2.20 | — | 1M | — | ✗ | ✓ | Legacy |
| MiniMax-01 LEGACY | $0.2 | $1.10 | — | 1M | — | ✗ | ✓ | Legacy |
Pricing across the lineup.
The MiniMax story so far.
Releases, the IPO, and the media-gen business. Sourced from MiniMax docs/press, the HKEX listing, and Wikipedia — verified at publication.
platform.minimax.io — the managed API for the M-series text models, with pay-as-you-go pricing and prompt caching (cache reads as low as $0.03/M).
OPEN-WEIGHT · SELF-HOSTMiniMax open-sources select models — MiniMax-M1 and MiniMax-Text-01 weights are published for self-hosting, including the 1M-context variants.
CONSUMER · MEDIAHailuo AI (video / audio / music generation) and Talkie (companion chat) are MiniMax's consumer apps — the media-gen heritage that sets it apart from text-only labs.
MiniMax (稀宇科技) was registered in Shanghai in December 2021 by Yan Junjie, who previously spent more than six years in research leadership at SenseTime. It is one of China's "Six Little Tigers" of AI, backed by Alibaba, Tencent, and HongShan.
MiniMax is unusual for spanning both text models and consumer media generation. On the text side, the agentic M2 family (M2.5, M2.7) offers cheap reasoning at $0.30/$1.20 with very low cache-read rates, and MiniMax M1 (June 2025) is an open-weight model with a 1M-token context. On the media side, Hailuo generates video, audio, and music, and Talkie is a popular companion-chat app launched internationally in 2023.
The company listed in Hong Kong on January 9, 2026, with shares jumping 109% on debut to a market cap above ~$12.8 billion — making Yan a billionaire and marking one of the first major "Six Tigers" IPOs. MiniMax also open-sources select models (MiniMax-M1, MiniMax-Text-01).
One weakness to state plainly: in September 2025, Disney, Universal, and Warner Bros. Discovery sued MiniMax in the US, alleging its AI-generated videos and images reproduced copyrighted studio characters — a real legal overhang on the media-gen side of the business.
The trade-offs: the M-series text models are text-only (multimodality lives in the separate Hailuo product), the company is China-based with data stored under Chinese rules on the direct API, and US copyright litigation and export-control questions apply. Versus DeepSeek and Moonshot, MiniMax's distinctive angle is pairing cheap agentic text with a real video/audio generation business.
DeepSeek
MIT open weights at the lowest cost-per-token for reasoning. MiniMax's M2 is similarly cheap; its differentiator is the Hailuo media-gen business, not raw text cost.
CHINESE PEERAlibaba (Qwen)
Broad open-weight catalog with full multimodality in one family. MiniMax splits text (M-series) and media (Hailuo) into separate products.
SIX-TIGERS PEERMoonshot (Kimi)
1T-param open-weight MoE and long-context heritage. Both are Tigers chasing agentic use; MiniMax adds video/audio generation Moonshot doesn't.
SIX-TIGERS PEERZhipu (GLM)
GLM family with free tiers and strong coding integration. Zhipu leans coding/agentic text; MiniMax leans agentic text plus consumer media.
PRIMARY RIVALOpenAI
Frontier benchmarks + Sora video and a global ecosystem. OpenAI competes with MiniMax on both text and video, but at far higher text prices.
CHINA CONSUMER LEADERByteDance (Doubao)
The Doubao model family from ByteDance — parent of TikTok and Douyin. Served through the Volcano Ark (火山方舟) platform on ByteDance's Volcano Engine cloud, Doubao powers China's most-used consumer AI app and undercuts most frontier labs on price.
Frequently asked.
Practical questions about MiniMax's models, media gen, and data residency.
Q · 01 Which MiniMax model should I start with? +
$0.30/$1.20, $0.06/M cache reads) is the current frontier pick with 1M context and multimodal input. M2.7 remains at the same standard token price with a 205K context. See the use case picker above.