MiniMax Changelog

2026-06

MiniMax-M2 language model

MiniMax continues to develop its M2 language model family, positioned as a general-purpose model with strong Chinese and English support. It is available through the MiniMax API and is increasingly used for agent workflows in Asia-Pacific markets.

Hailuo 2.3 video generation

Hailuo 2.3 improves complex video performance and introduces tighter media-agent integration. The headline feature is better motion consistency, character coherence, and camera control across longer clips.

S2V-01 subject reference

Released earlier in the year, S2V-01 adds subject-reference capabilities to Hailuo. You can provide a reference image of a person, object, or character and generate video that keeps the subject visually consistent across scenes.

2026-05

API platform expansion

MiniMax expanded its API documentation and release notes hub, covering language, audio, video, image, and music models. Token plans and pricing tiers were adjusted to compete with other Chinese model providers.

What this means for users

MiniMax is best known in the West for Hailuo video generation, but its language models are gaining traction for cost-sensitive agent workloads. If your use case involves video + language together — summarizing clips, generating scripts, or producing social content — MiniMax is one of the few integrated stacks worth evaluating.

What to watch

• Independent video quality comparisons against Runway, Kling, and Luma.
• Whether MiniMax improves API latency and reliability for non-China developers.
• Pricing pressure as video-generation competition intensifies.

Source: MiniMax release notes