7 MiniMax models for agentic apps

This list shows seven MiniMax models and what each one is best suited for.

MiniMax’s release notes now span text, speech, video, image, and music, with MiniMax-M3 arriving on Jun. 1, 2026 as the latest M-series model for agentic reasoning, tool use, coding, multimodal chat input, and long-context work.

Item	Release date	Main use
MiniMax-M3	Jun. 1, 2026	Agentic reasoning, coding, multimodal chat, long context
MiniMax-M2.7	Mar. 18, 2026	Reasoning and tool use
MiniMax-M2.5	Feb. 2026	Programming, tool calling, search, office tasks
Speech-2.8	Jan. 23, 2026	Natural voice and audio quality
Music-2.5+	Mar. 2026	Instrumental music generation
MiniMax-Hailuo-2.3	Oct. 28, 2025	Video generation with realism and prompt adherence
Image-01	Feb. 15, 2025	Text-to-image generation

1. MiniMax-M3

MiniMax-M3 is the newest M-series language model in the release notes, and it is the one to watch if you are building agentic apps. MiniMax says it is meant for reasoning, tool use, coding, multimodal chat input, and long-context tasks.

If your product needs a general-purpose model that can think, call tools, and handle mixed inputs, M3 is the cleanest starting point in this set.

• Best fit: coding assistants
• Best fit: tool-using agents
• Best fit: long-document workflows

2. MiniMax-M2.7

MiniMax-M2.7 arrived on Mar. 18, 2026 as a follow-up M-series release. The notes describe the M2.7 series as aimed at recursive self-improvement, which signals a focus on iterative reasoning and stronger agent behavior.

Choose M2.7 if you want an M-series option that sits close to the newest release but may be a better fit for experimentation, comparison testing, or staged upgrades.

• Series: MiniMax-M2.7 / M2.7-highspeed
• Release note focus: recursive self-improvement
• Good for: agent loops and eval runs

3. MiniMax-M2.5

MiniMax-M2.5 landed in Feb. 2026 and is pitched around practical work. MiniMax says the M2.5 series targets programming, tool calling, search, office productivity, and other scenarios where a model needs to do useful work, not just chat.

This is the model family to consider when you need broad utility across business tasks and developer workflows, especially if your app mixes retrieval, automation, and text generation.

• Series: MiniMax-M2.5 / M2.5-highspeed
• Named strengths: programming and search
• Also aimed at: office productivity

4. Speech-2.8

Speech-2.8 was released on Jan. 23, 2026 and focuses on voice quality. The release note highlights natural sound tags, lifelike voice, and pristine audio quality, which makes it a strong pick for speech products that need a more human feel.

For assistants, narrators, and voice interfaces, Speech-2.8 is the most direct choice in the list because its goals are easy to map to product needs.

• Release note cues: natural sound tags
• Audio goal: lifelike voice
• Use case: assistants and narration

5. Music-2.5+

Music-2.5+ was released in Mar. 2026 with a clear creative angle. MiniMax describes it as an instrumental-focused model that helps break through style boundaries, which points to stronger control for music generation.

If your workflow is about composing background tracks, experimenting with genre, or generating instrumental pieces for content, this model is the most relevant music release in the set.

• Release note phrase: instrumental unlocked
• Primary output: music
• Good for: content teams and creators

6. MiniMax-Hailuo-2.3

MiniMax-Hailuo-2.3 and 2.3-Fast were released on Oct. 28, 2025 for video generation. MiniMax highlights body movement, facial expressions, physical realism, and prompt adherence, which are the details that matter when video output needs to feel believable.

This is the pick for teams building short-form video tools, product demos, or visual storytelling apps where motion quality matters as much as prompt fidelity.

• Variants: 2.3 and 2.3-Fast
• Core strengths: realism and expression
• Output type: video

7. Image-01

Image-01 was released on Feb. 15, 2025 and supports text-to-image generation in multiple sizes. It is the simplest visual model in this list, and that makes it useful when you need quick image generation without adding video or audio complexity.

Image-01 is a good entry point for design mockups, illustration concepts, and app features that turn prompts into visuals.

• Release date: Feb. 15, 2025
• Input type: text prompt
• Output type: image

How to decide

If you are building an agent, start with MiniMax-M3. If you want a slightly earlier M-series option for testing or comparison, look at M2.7 or M2.5. For voice, Speech-2.8 is the clearest fit. For creative media, use Music-2.5+ for audio, Hailuo-2.3 for video, and Image-01 for still images.

The fastest way to choose is to match the model to the output format first, then narrow by task depth. Text and agents point to the M-series, speech points to Speech, and media generation points to Music, Hailuo, or Image.