MiniMax

MiniMax

Overview

MiniMax is a Shanghai-based Chinese AI company that develops a broad portfolio of multimodal foundation models spanning text, audio, image, video, and music — and ships them through consumer-facing AI-native products including Hailuo (video generation), Talkie (companion chat), and others. Founded in late 2021, MiniMax has emerged as one of the most prolific Chinese AI labs by product output, with a 2026 portfolio that spans M-series text models, the Hailuo 2.x video generation family, the Music 2.x music generation models, and an advanced TTS stack.

Hailuo, the company's video generation model, has consistently ranked at or near the top of independent video benchmarks. Hailuo 02 reached #2 globally on Artificial Analysis benchmarks (surpassing Google Veo 3 but trailing Seedance 1.0), and the recent Hailuo 2.3 release further extended performance on dynamic motion, character micro-expressions, and physical action portrayal. MiniMax's positioning emphasizes consistency across modalities and a vertically integrated product strategy — not just APIs, but consumer products built on top of the company's own models.

Key Details

  • Founded: 2021
  • Headquarters: Shanghai, China
  • Distribution Model: Consumer apps (Hailuo, Talkie) + APIs (MiniMax developer platform, partner integrations)
  • Website: https://www.minimax.io / https://hailuoai.video

Current Models

  • [[Hailuo 2.3]] — Latest video generation model; significantly improved physical action portrayal, stylization, and character micro-expressions; subject consistency across 6–10 second clips; four variants (Standard, Pro, Fast, Fast Pro)
  • Hailuo 02 — Previous-generation video model; ranked #2 globally on Artificial Analysis at release, surpassing Veo 3
  • [[MiniMax M2.7]] — Open-weight self-evolving agentic model released March 18, 2026. 230B parameter MoE (10B active per token); 205K context; SOTA among open-weight on SWE-Pro (56.22%) and Terminal Bench 2 (57.0%); top GDPval-AA among open weights (ELO 1495). Released on Hugging Face under a non-commercial license (commercial use requires written authorization).
  • MiniMax Music 2.6 — Music generation model
  • MiniMax TTS — Advanced TTS stack with consumer and developer-platform availability

Recent Developments

  • MiniMax M2.7 Open-Weight Release (March 18, 2026): MiniMax open-sourced M2.7 — a 230B-parameter MoE (10B active) "self-evolving" agentic model — on Hugging Face. SWE-Pro 56.22% (matching GPT-5.3-Codex), Terminal Bench 2 57.0%, GDPval-AA ELO 1495 (highest among open weights). NVIDIA highlighted M2.7 in its developer blog as a flagship for scalable agentic workflows. M2.7 is one of four Chinese open-weight coding models (with Z.ai GLM-5.1, Moonshot Kimi K2.6, DeepSeek V4) released within a 12-day window in spring 2026.
  • Hailuo 2.3 Release (Spring 2026): Significant improvements in physical action portrayal, stylization, and character micro-expression rendering. Four variants (Standard, Pro, Fast, Fast Pro) with up to 50% cost reduction in batch creation via the Fast variant.
  • MiniMax 2026 Portfolio: M-2.7 text model, Music 2.6, advanced TTS, and Hailuo 2.3 — positioning MiniMax as a multimodal foundation model provider rather than a single-modality specialist.
  • AI Distillation: 2026 reporting indicates MiniMax has emphasized model distillation techniques — compressing larger models into faster, cheaper variants — as a core operational strategy, particularly for the Hailuo Fast variants.
  • Benchmarks: Hailuo 02's #2 global ranking on Artificial Analysis (above Veo 3) at release was a notable validation point for non-Western video generation models.

Why They Matter

MiniMax is the most multimodal-portfolio-complete Chinese AI lab — competitive across text, audio, image, video, and music in parallel rather than focused on a single modality. The Hailuo line specifically has consistently ranked among the top global video models on independent benchmarks, and the four-variant pricing structure (Standard/Pro/Fast/Fast Pro) explicitly addresses the cost dimension that limits production-scale AI video adoption. Strategically, MiniMax sits in a different position than Kuaishou (also Chinese, also strong in video): where Kuaishou translates a consumer-platform distribution moat into video AI leadership, MiniMax operates as a model-portfolio company shipping across modalities, with Hailuo as one of several frontier products rather than the singular flagship. For Western enterprises that need cost-efficient video generation and can navigate the geopolitical considerations around Chinese-origin AI models, Hailuo 2.3 — particularly the Fast variant — is one of the strongest cost-quality-points on the market.

Last Updated

May 9, 2026

→ Back to Companies