MiniMax

Name

MiniMax

Legal name

Shanghai Xiyu Technology Co., Ltd. (上海稀宇科技有限公司)

Website

https://minimaxi.com

Company type

Public

Founded year

2022

Operating status

Operating

Headcount range

251–500 employees

Ownership category

akta.pro rank

Product category

AI Foundation Models

NAICS

Software Publishers (5132)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Multimodal Generation (Text-Image-Video-Audio) & Creative Toolchains (HDAAACAG), Fine-Tuning, Adaptation & Custom Model Training (PEFT/LoRA/RLHF) (HDAAACAC), AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ)

Keywords

• Multimodal AI models
• Large language models
• AI video generation
• AI coding agents
• Foundation model platform

Headquarters

HQ city

Singapore

HQ country

Singapore

HQ region

Asia

Offices1 record

Markets served

GTM type

B2B and B2C

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations

Revenue model

1. API and Token Plan Subscriptions: Recurring subscription revenue via Plus (¥49/mo), Max (¥119/mo), and Ultra (¥469/mo) Token Plan tiers, plus per-token usage-based API billing on platform.minimaxi.com for individual developers and enterprises.
2. Enterprise and Developer API Consumption: Usage-based API revenue from 1M+ enterprise customers and developers across 100+ countries; reported 2025 revenue of approximately $79M with 70% from international markets.
3. Consumer AI Product Subscriptions and Freemium: Free-tier and paid usage of consumer products Hailuo AI (video), Xingye (interactive entertainment), MiniMax Audio (speech/music), and MiniMax Code (agent). Includes free trial allocations such as Music 2.6's 500 songs/day per account.
4. Open-Weight Commercial Licensing (M3): MiniMax Community License for the open-weights M3 release with a separate commercial licensing agreement required for revenue-generating deployments; HuggingFace-hosted weights plus hosted API.
5. Platform Partner Revenue Share: Ecosystem revenue share through cloud partner integrations including Huawei Cloud INSPIRE Model Partner Program, Dell Enterprise Hub, and OpenRouter distribution.

Pricing tiers

Go-to-market motion6 records

Distribution channels8 records

Marketing channels8 records

Core offering

MiniMax develops and sells self-developed multimodal AI foundation models (MiniMax M3/M2.7/M2.5 language models, Hailuo 2.3 video, Speech 2.8 voice, Music 2.6) and consumer AI applications. Distribution occurs through the MiniMax Open Platform (REST APIs, Token Plan subscriptions at ¥49/¥119/¥469 monthly, Coding Plan) for developers and enterprises, alongside consumer apps (Hailuo AI video, MiniMax Audio, Xingye character AI, MiniMax Code) reaching 300M+ users globally.

Product overview

MiniMax operates a platform-plus-modules architecture centered on self-developed multimodal foundation models and a suite of consumer AI-native applications. The foundation model layer comprises the flagship language model line (MiniMax M3, M2.7, M2.5) built on the proprietary MSA sparse attention architecture, the Hailuo 2.3 / 2.3 Fast video generation model, MiniMax Speech 2.8 text-to-speech model, and MiniMax Music 2.6 music generation model. These models are exposed to developers through the MiniMax Open Platform (platform.minimaxi.com) via REST APIs and to end users through consumer applications including Hailuo AI (海螺视频) for video, MiniMax Audio for speech and music, Xingye (星野) for AI character/role-play, MiniMax Hub, MiniMax Code for AI coding, and MiniMax Agent / Mavis for long-horizon general agentic tasks. The MiniMax Token Plan and Coding Plan subscriptions distribute model access across third-party AI Coding Tools (Claude Code, Cursor, OpenCode, etc.), while a separate MMX-CLI command-line interface lets AI agents invoke seven generative modalities directly. The company markets itself as a globally leading general AI technology firm serving over 300 million individual users across more than 200 countries and regions, and over 1 million enterprise customers and developers across more than 100 countries.

Differentiator

Problem solved

Functional benefit

Brands

• Hailuo AI (海螺AI): AI video generation platform including Hailuo 2.3 video model and Media Agent for multimodal content creation.
• MiniMax Code
• Xingye (星野)
• MiniMax Audio / Talkie
• MiniMax Hub

Products and services

• MiniMax M3 Frontier flagship large language model using self-developed MSA sparse attention architecture, supporting up to 1M token context window with native multimodality. Targets enterprise developers and coding agent vendors with frontier-grade SWE-Bench Pro performance at 5-10% of US frontier model cost.
• MiniMax M2.7 Self-evolving 230B-parameter MoE language model that drives its own RL iteration loop, supports Agent Teams and Tool Search Tool, and delivers frontier agentic performance for coding and software engineering tasks.
• MiniMax M2.5 Language model for Agent Universe workloads with SWE-Bench Verified 80.2% and BrowseComp 76.3%. Available in 100 TPS and 50 TPS variants at 1/10 to 1/20 the cost of Opus/Gemini/GPT-5; open-sourced on HuggingFace.
• MiniMax Hailuo 2.3 / 2.3 Fast Video generation model with enhanced physical performance, anime/illustration/ink/game-CG style support, facial micro-expression control, and optimized object motion response. Fast variant delivers 50% lower cost for batch creation. Powers the Hailuo AI consumer application.
• MiniMax Speech 2.8 Text-to-speech voice large model with native filler-word modeling, 10-second voice cloning, studio-grade noise reduction, and 40+ language cross-lingual synthesis. Powers the MiniMax Audio consumer product.
• MiniMax Music 2.6 Music generation model with BPM/Key/structure-controllable prompts, Cover feature for melody-preserving style swap, and first-packet latency under 20 seconds. Open-sourced Music Skills include minimax-music-gen, minimax-music-playlist, and buddy-sings.
• Hailuo AI (海螺AI) Consumer-facing web and mobile application providing access to Hailuo video generation and the Media Agent for one-click multimodal video creation with image, video, and audio inputs.
• MiniMax Audio Consumer-facing speech and music generation web application powered by MiniMax Speech 2.8 and Music 2.6, offering voice cloning, multilingual TTS, and music creation with Cover feature for individual creators.
• Xingye (星野) Consumer-facing AI character and role-play product featuring the MiniMax-M2-her Role-Play Agent with persistent persona and dialogue capability; serves individual entertainment seekers.
• MiniMax Code General-purpose Agent platform and coding agent designed for MiniMax-M3 with Agent Team multi-agent orchestration, Producer+Verifier adversarial Harness, persistent memory, and Computer Use capability. Built on OpenCode and Pi Agent; available as a desktop app.
• MiniMax Agent / Mavis Long-horizon general Agent platform with built-in MiniMax MCP, Office Skills (Word/PPT/Excel), GitHub/Gitlab/Slack/Figma MCP integrations, and agent-native multimodal output. Rebranded as Mavis with Agent Team capabilities for continuous multi-day task execution.
• MiniMax Open Platform Developer-facing open platform at platform.minimaxi.com providing API documentation, Token Plan subscriptions, API pricing, user center, and service status for enterprise and developer customers worldwide.
• MiniMax Token Plan Monthly subscription access to MiniMax frontier models (M3, M2.7) for AI Coding Tools (Claude Code, Roo Code, Kilo Code, Cline, Cursor, OpenCode, Codex CLI, Droid, TRAE, Grok CLI). Three tiers: Plus ¥49 (600M tokens), Max ¥119 (1.8B tokens), Ultra ¥469 (5.5B tokens).
• MiniMax Coding Plan Subscription access to MiniMax M2.5 for AI coding workflows with 100 TPS standard and highspeed variants; pricing at 1/10 to 1/20 of comparable frontier models.

Quantifiable outcome

• M3 achieves SWE-Bench Pro 59.0% and exceeds Opus 4.7 on SVG-Bench
• +9 more outcomes

Named customers10 records

Segments6 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration17 records

AI capability15 records

Feature10 records

Partnerships

16 partnerships are on record, tiered major, minor and flagship.

• Cast AImajorStrategic or Co-development Partner · 19 June 2026Cast AI integrated MiniMax M3 as the default builder model in its Kimchi autonomous coding agent — first coding agent to default to an open-weight frontier model. Achieved 2.5x cost reduction and 59% SWE-bench Pro score. Deployed across AWS, GCP, Azure, on-premises, and air-gapped environments.
• Shanghai International Film Festival (SIFF) - AI Studio programminorGTM or Marketing Partner · 17 June 2026MiniMax's Hailuo AI content-generation platform partnered with SIFF to launch the festival's first AI-focused program 'AI Studio' — a month-long collaborative program training 22 selected filmmakers in AI-assisted video creation.
• Huawei Cloud (Huawei Ascend)flagshipTechnology or Integration · 16 June 2026On June 12, 2026, Huawei Cloud completed Day-0 adaptation of MiniMax M3 model on its CloudMatrix service powered by Ascend compute. Huawei provides Tokens-level compute support with MSA operator adaptation and MOE balancing optimization. Co-launched the 'Hundreds of Models, Thousands of Forms' ecosystem partnership with 20+ TOP model vendors at Huawei Cloud INSPIRE 2026.
• Biren Technology (壁仞科技)flagshipTechnology or Integration · 16 June 2026Biren's Biren Sui (壁砺) 166 series GPUs completed Day-0 adaptation of MiniMax M3 via SGLang inference framework on June 12, 2026. Previously adapted MiniMax M2.1, M2.5, and M2.7 models with strong model precision and stability. Biren has now adapted 20+ leading large models.
• Haiguang Information (海光 DCU)flagshipTechnology or Integration · 16 June 2026Haiguang DCU completed Day-0 full-flow adaptation of MiniMax M3 on June 12, 2026. Provides zero-modification out-of-the-box deployment with architecture-level optimization through proprietary DAS/DTK software stack. Enables national-sovereign AI deployments.
• Moore Threads (摩尔线程)flagshipTechnology or Integration · 16 June 2026Moore Threads' MTT S5000 AI compute card completed Day-0 adaptation of MiniMax M3 on June 12, 2026. Provides native FP8 acceleration, 80GB VRAM, and 1.6TB/s bandwidth. Supports vLLM and SGLang deployment frameworks via MUSA software stack.
• MetaX (沐曦股份)flagshipTechnology or Integration · 16 June 2026MetaX's Xiyun C-series GPU completed Day-0 deep adaptation of MiniMax M3 on June 12, 2026 via MXMACA full-stack software stack. MetaX has completed Day-0 adaptation for 26 models including all MiniMax M-series flagship models.
• Kunlunxin (昆仑芯)flagshipTechnology or Integration · 16 June 2026Kunlunxin completed Day-0 adaptation of MiniMax M3 on June 12, 2026, becoming one of the first Chinese compute vendors to support the model. Part of ongoing Day-0 partnership across all MiniMax flagship models.
• SAIHEAT Limited (NASDAQ: SAIH)minorChannel Partner/ Reseller/ Distributor · 11 June 2026SAIHEAT announced strategic expansion into AI inference services, providing enterprise-level authorized token access to mainstream open-source AI models including Kimi, GLM, DeepSeek, MiniMax, and MiMo. Targets enterprise customers seeking to deploy AI workloads without managing GPU clusters.
• Maison Solutions + SupplyAiminorStrategic or Co-development Partner · 9 June 2026Maison Solutions Inc. announced strategic collaboration with SupplyAi and MiniMax to explore AI-native solutions for the North American food supply chain — workflow automation, operational visibility, and decision-making. No specific financial terms or implementation timeline disclosed.
• Huawei Cloud INSPIRE 2026 Ecosystem ProgramflagshipGTM or Marketing Partner · 9 June 2026MiniMax joined Huawei Cloud's AI Model Partner Program with 20+ model providers at INSPIRE 2026 in Shanghai. Includes Smart Healthcare, Embodied AI, Smart Manufacturing, and Scientific Computing Industry Zones. Targets 5,500+ Huawei Cloud enterprise customers globally.
• Alipay / Ant GroupflagshipStrategic or Co-development Partner · 26 May 2026Alipay partnered with MiniMax and Stepfun to launch AI Wallet and Token Pay, the world's first AI-native payment infrastructure for agentic commerce. MiniMax M3 integrated Alipay TokenPay enabling voice-command checkout via Mix Claw service. Alipay AI Pay has reached 100M users and 300M transactions.
• Dell TechnologiesflagshipOEM/ Whitelabel/ Licensing Partner · 18 May 2026Dell added five open-weight models including MiniMax M2.7 to its Dell Enterprise Hub platform optimized for PowerEdge XE9780 servers with 8x NVIDIA HGX B300 GPUs and 288GB HBM3 memory. Part of broader industry shift toward open-source frontier models coexisting in multi-agent long-context workflows.
• ThinkingAI (ThinkingData)majorStrategic or Co-development Partner · 17 April 2026ThinkingAI partnered with MiniMax to launch Agentic Engine, an enterprise agentic platform deployed within customer infrastructure using MiniMax's multimodal foundation models. Serves 1,500+ companies including Sega, Krafton, Century Games, and Habby. Co-founder Chris Han unveiled the platform at Computer History Museum in Mountain View.
• Huawei Ascend (昇腾)flagshipTechnology or Integration · 12 April 2026Huawei Ascend AI compute hardware and vllm-Ascend inference engine completed Day-0 adaptation of MiniMax M2.7. Supports Atlas 800 A3 and Atlas 800I A2 product series. Includes FlashComm sequence splitting, all-reduce to ReduceScatter/AllGather acceleration, and MoE fused operators.
• FlagOS Community (众智 FlagOS)majorTechnology or Integration · 12 April 2026FlagOS community completed Day-0 multi-chip adaptation of MiniMax M2.7 across 6+ AI chips (Pingtougue, Huawei, Haiguang, MetaX, Tianshu, NVIDIA) via unified vLLM-plugin-FL and FlagGems operator library. Provides zero-code-change cross-chip migration in 36 hours.

Scale indicators13 records

Recent moves8 records

Expansion highlights8 records

Direct peers

• DeepSeek: Hangzhou-based Chinese AI lab shipping frontier open-weight LLMs (V3, V4 series) that lead MiniMax in OpenRouter token consumption. Direct competitor in cost-efficient open-weight foundation models for coding and reasoning, with comparable developer adoption.
• Zhipu AI: Beijing-based Chinese AI lab (GLM series) also pursuing a Hong Kong listing and developing multimodal foundation models with comparable enterprise and developer traction. Bank of America and JPMorgan cover both as peer Chinese AI model labs.
• Moonshot AI: Beijing-based Chinese AI lab behind the Kimi models, named alongside MiniMax in Anthropic's distillation accusations and US House investigation. Direct competitor in long-context and agentic Chinese LLMs.
• Stepfun: Shanghai-based Chinese multimodal AI lab developing foundation models across text, image, and video. Co-partner with MiniMax in Alipay's Agentic Commerce Trust Protocol, indicating direct overlap in multimodal AI and agentic commerce.

Broad incumbents

• Alibaba (Qwen): Qwen team within Alibaba develops open-weight foundation models (Qwen 3, Qwen-VL) with substantial cloud distribution via Alibaba Cloud. A strategic shareholder in MiniMax and a direct competitor in open-weight LLMs with significantly larger infrastructure footprint.
• ByteDance (Doubao): ByteDance operates the Doubao LLM family and Volcano Engine cloud, with major video generation models (e.g., Seedance). Forbes listed ByteDance alongside MiniMax as a Chinese AI lab that overtook OpenAI in video generation, making it a broad incumbent competitor.
• OpenAI: US frontier AI lab behind GPT-5 and ChatGPT. M3's published benchmarks are positioned against GPT-5.5/Opus 4.7; OpenAI's enterprise and developer API ecosystem is the primary incumbent MiniMax targets with its cost advantage.
• Anthropic: US AI lab behind Claude (Opus 4.7) and the Claude Code agentic coding tool that MiniMax M3 directly integrates with via Token Plan. Anthropic's distillation accusations against MiniMax make it both a competitor and a primary antagonist in the IP dispute.

Emerging players

• Baichuan Intelligence: Beijing-based Chinese AI lab founded by the former Sogou CEO, focused on open-weight foundation models. Part of the broader Chinese AI lab cohort competing for similar cost-efficient enterprise and developer workloads.
• 01.AI (Yi): Beijing-based AI lab founded by Kai-Fu Lee shipping the Yi family of open-weight LLMs. Operates in adjacent Chinese open-weight LLM segment, with smaller scale and traction than MiniMax but overlapping customer segments.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks7 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/MiniMax-AI
• LinkedInlinkedin.com/company/minimax-ai

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds9 records

Investors12 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A1 record

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about MiniMax