MiniMax

[Shanghai, China][founded 2022][backed by Alibaba; Tencent; listed on HKEX (100) since January 2026]www.minimax.io

Founded in early 2022 and headquartered in Shanghai, MiniMax develops foundation models for text (the M series), video (Hailuo and H3), speech and music; it listed on the Hong Kong Stock Exchange in January 2026 with Alibaba and Tencent among its investors. MiniMax M3 (June 2026) is a 428B open-weight MoE with a 1M-token context, and MiniMax H3 (July 2026) is an open multimodal video model. Its consumer products include the Hailuo video app and the Talkie character app.

sources:en.wikipedia.org minimax.io platform.minimax.io

· verified 2026-09-13

models

families

open weights

3d teardowns

2025–2026

shipping

■ video 11■ text 1■ audio 1

13 models · 3 families · 2025–2026

MiniMax M series (text)

newest: MiniMax M3 · 2026-05-31

[open-weights][api]

Hailuo and H3 (video)

newest: MiniMax H3 Max Lip Sync · 2026-09-17

[video][open-weights][api][H3 / H3 Max / H3 Max Turbo]

MiniMax Music (audio)

newest: MiniMax Music · 2025-01

[audio][api]

Families (3)

MiniMax M series (text)

[open-weights][api]

1 in catalog · 3 earlier

before the catalog2025-01 MiniMax-Text-01, first model of the line→2025-10 MiniMax M2, agentic model, first M2 release→2026-03 MiniMax M2.7, recursive self-improvement focus→

MiniMax M3

2026-05-31

428B MoE (23B active), 1M context, native multimodal input

[text]1M ctx[open-weights · MiniMax Community License][3d teardown]newest

sources:platform.minimax.io huggingface.co

Hailuo and H3 (video)

[video][open-weights][api][H3 / H3 Max / H3 Max Turbo]

11 in catalog · 2 earlier

before the catalog2025-06 Hailuo 02, 1080p, 10-second clips→2025-10 Hailuo 2.3, body movement and facial expression→

MiniMax H3

2026-07-31

33B omni-modal video model, 2K and 4K, native audio, open weights

[video][3 endpoints]

MiniMax H3 Max

2026-09

fal post-trained variant of H3 for prompt adherence

[video][5 endpoints]

MiniMax H3 Max Director

2026-09

Continuous realtime video streamed over WebRTC and steered with live prompts and scripted beats, keeping characters and settings consistent across a session of up to 15 minutes. Not available as a workflow node.

[video]