MiniMax began with the abab language models in 2022. Its first speech model arrived in 2023, followed by Video-01 and Music-01 in 2024. These early releases established MiniMax’s development in language, speech, video, and music generation.
The company entered a new phase in 2025 with MiniMax-01’s long-context language and vision-language models and M1’s advanced reasoning capabilities. The M2 and M3 generations further developed coding, AI agent tasks, and multimodal understanding. Meanwhile, Hailuo, Speech, and Music continued to evolve through new model generations.
The MiniMax AI model timeline traces this history from the earliest abab releases to M3.1-Flash-Preview in 2026. Follow the major milestones to see when each model family emerged, which releases introduced significant new capabilities, and how MiniMax’s language and media models evolved over the years.
Last updated: October 8, 2026.
MiniMax model timeline
2026
| Date | Milestone |
|---|---|
| September 27, 2026 | MiniMax-M3.1-Flash-Preview enters preview on MiniMax Code and subscription access, focusing on coding and multi-step development tasks. It accepts text, image, and video input and has a 1-million-token context window. |
| August 20, 2026 | Music API access changes Paid Music Generation and Lyrics Generation APIs close to new customers; existing paying customers retain access. Free music-generation APIs end, and music models are removed from M Plan. |
| August 13, 2026 | MiniMax-Music-3.0 is publicly introduced with downloadable model weights for complete songs up to five minutes, including vocals and accompaniment. |
| August 3, 2026 | MiniMax-H3 open weights are released as downloadable H3-Base checkpoints for first-and-last-frame and reference-conditioned video generation with audio. |
| July 31, 2026 | MiniMax-H3 launches for video generation with native stereo audio, multimodal input, up to 2K output, and clips up to 15 seconds. |
| July 16, 2026 | MiniMax-Music-3.0 API becomes available for complete-song generation, ahead of the model’s August open-weight announcement. |
| June 1, 2026 | MiniMax-M3 launches with open weights, MiniMax Sparse Attention, a 1-million-token context window, native image and video input, and coding and computer-use capabilities. |
| April 12, 2026 | MiniMax-M2.7 open weights are released for local deployment, following the March model launch. |
| April 10, 2026 | MiniMax-Music-2.6 launches in a global creative beta. It improves prompt control and low-frequency sound and introduces music generation from reference audio. |
| March 18, 2026 | MiniMax-M2.7 launches with stronger software-engineering and professional-task performance, native agent teams, and model-assisted self-improvement. |
| March 4, 2026 | MiniMax-Music-2.5+ introduces dedicated instrumental generation to the Music 2.5 family. |
| February 12, 2026 | MiniMax-M2.5 and M2.5-Lightning launch for coding, tool use, search, and office tasks. The Lightning variant provides higher output speed. |
| January 28, 2026 | MiniMax-Music-2.5 launches with section-level song structure control, expanded instrumentation, and more expressive vocals. |
| January 27, 2026 | MiniMax-M2-her technical overview is published for the roleplay-focused model, detailing character consistency and long conversations. MiniMax had reported an M2-her update in the fourth quarter of 2025. |
| January 23, 2026 | MiniMax-Speech-2.8 launches with native sound tags for breaths and fillers, alongside voice-cloning and expressive-speech improvements. |
2025
| Date | Milestone |
|---|---|
| December 23, 2025 | MiniMax-M2.1 launches with improved coding across Rust, Java, Go, and other languages and stronger engineering and office tasks. |
| October 31, 2025 | MiniMax-Music-2.0 launches with songs up to five minutes and improved vocal expression and musical arrangements. |
| October 30, 2025 | MiniMax-Speech-2.6 launches with lower-latency speech, improved format handling, and Fluent LoRA for voice-cloning fluency. |
| October 28, 2025 | MiniMax-Hailuo-2.3 and Hailuo-2.3-Fast launch with improved motion and character expressions. The Fast variant focuses on faster, lower-cost image-to-video generation. |
| October 27, 2025 | MiniMax-M2 launches with open weights, emphasizing coding, agent execution, and tool use. |
| September 11, 2025 | MiniMax-Music-1.5 launches after a June beta, extending song generation to four minutes with more control over vocals and arrangement. |
| August 7, 2025 | MiniMax-Speech-2.5 launches with improved multilingual voice cloning and support for more than 40 languages. |
| June 18, 2025 | MiniMax-Hailuo-02 launches with native 1080p video generation, stronger instruction following, and improved physical motion. |
| June 16, 2025 | MiniMax-M1 launches with open weights, a hybrid attention architecture, and a 1-million-token context window for reasoning tasks. |
| April 2, 2025 | MiniMax-Speech-02 launches in HD and Turbo variants for multilingual text-to-speech and voice cloning. |
| March 3, 2025 | T2V-01-Director and I2V-01-Director launch for text-to-video and image-to-video generation with more explicit camera controls. Both are available through Hailuo AI and the API. |
| February 28, 2025 | MiniMax-Image-01 launches as MiniMax’s first text-to-image model, with API access available at release. |
| January 15, 2025 | MiniMax-Text-01 and MiniMax-VL-01 launch with open weights. The MiniMax-01 family introduces Lightning Attention and a long-context language model alongside a vision-language variant. |
| January 15, 2025 | MiniMax-Speech-01-HD launches with improved speech synthesis, voice cloning, and multilingual expression. |
| January 12, 2025 | Hailuo I2V-01-Live launches for animating 2D illustrations with more stable character motion and expression. |
| January 10, 2025 | Hailuo S2V-01 launches for subject-reference video generation from a single reference image. |
2024
| Date | Milestone |
|---|---|
| November 6, 2024 | abab7-preview is released through MiniMax’s developer platform, improving long-text, math, and writing tasks over the abab6.5 generation. |
| August 31, 2024 | Video-01 and Music-01 launch, establishing MiniMax’s video- and music-generation model families. Video-01 supports text-to-video and image-to-video generation. |
| April 17, 2024 | abab6.5 and abab6.5s launch as long-context mixture-of-experts language models with 200,000-token context windows. |
| January 2024 | abab6 launches as a mixture-of-experts text model in the abab language-model family. |
2023
| Date | Milestone |
|---|---|
| November 2023 | MiniMax-Speech-01 launches as the company’s first speech-generation model. |
| May 2023 | abab5.5 launches as MiniMax begins its first commercial API customer relationship. |
2022
| Date | Milestone |
|---|---|
| October 2022 | abab3 completes training as an early text-model iteration. |
| June 2022 | abab2 completes training as the second abab text-model iteration. |
| April 2022 | abab1 launches as MiniMax’s first text model. |
How MiniMax model families are organized
The abab, MiniMax-01, and M-series names identify successive language-model generations. MiniMax-VL-01 is the vision-language model in the MiniMax-01 family. M2-her focuses on character and roleplay conversations.
Hailuo and H-series models generate video. MiniMax Speech models produce spoken audio, and the Music series generates songs and instrumental music. MiniMax Code, Hailuo AI, and MiniMax Audio are products that provide access to selected models.
Related Resources
- AI Model Release Calendar: Latest AI Model Releases by Date
- Hailuo AI: AI Video Generator Powered by Minimax H3
