Overview
MiniMax M3.1-Flash-Preview is a preview model in MiniMax's M-series that debuted in MiniMax Code v3.0.74 on September 28, 2026. MiniMax's API documentation lists it as MiniMax-M3.1-Flash-Preview, a "Frontier multimodal coding model with 1M context window and tunable thinking depth", and the MiniMax Code changelog describes it as supporting everyday development work from bug fixes to complete feature implementation.
The model accepts text, image and video input with a context window of up to 1,000,000 tokens, aimed at long documents, codebases and multi-step agent sessions. Reasoning is always on: instead of a thinking on/off switch as on MiniMax M3, depth is tuned with an effort parameter that takes low, medium, high, xhigh or max, with max as MiniMax's recommended default. MiniMax's preview guide lists a Mixture-of-Experts architecture with 428B total and about 23B active parameters, sparse attention and a native visual encoder, at roughly 150 output tokens per second depending on load.
At release, MiniMax made the preview available only through its M Plan subscription and inside MiniMax Code, reachable from the API with a subscription key. It published no pay-as-you-go price, no benchmark table and no open weights for the preview.
| Released | 2026-09-28 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | 428B total / ~23B active (per MiniMax's preview guide) |
| Context | 1M |
| Architecture | Mixture-of-Experts with sparse attention and a native visual encoder |
| Modalities | Text, Vision, Video |
| Status | Preview — available only through MiniMax's M Plan (Token Plan subscription) and MiniMax Code; not offered pay-as-you-go, no published per-token price and no weights released |
Strengths
- 1,000,000-token context window for whole codebases and long agent sessions
- Native text, image and video input
- Five-level reasoning effort control (low, medium, high, xhigh, max)
- Built for agentic reasoning, tool use, coding and structured task execution
Best for
- Everyday software development, from bug fixes to complete feature implementation
- Agentic coding sessions in MiniMax Code
- Frontend and 3D web work from text and visual references
- Long-document and repository understanding
How to access
| Provider | Model ID |
|---|---|
| MiniMax (M Plan subscription key) ↗ | MiniMax-M3.1-Flash-Preview |
MiniMax M-Series — every version
The full lineage of the MiniMax M-Series line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| MiniMax M3.1-Flash-Preview | 2026-09-28 | 1M | Proprietary |
| MiniMax M3current | 2026-06-01 | 1M | MiniMax Community |
| MiniMax M2.7 / M2.7-highspeed | 2026-03-18 | — | Open weights |
| MiniMax M2.5 / M2.5-Lightning | 2026-02-12 | — | Open weights |
| MiniMax M2.1 | 2025-12-23 | — | Open weights |
| MiniMax M2 | 2025-10-27 | — | MIT |
FAQ
When was MiniMax M3.1-Flash-Preview released?
It debuted in MiniMax Code v3.0.74, dated September 28, 2026 in MiniMax's changelog. MiniMax's September 29, 2026 subscription announcement and the MiniMax Code v3.1.0 release on September 30, 2026 followed, the latter adding unlimited M3.1-Flash-Preview use for subscribers from October 1 to October 7.
Can I use M3.1-Flash-Preview through the pay-as-you-go API?
No. MiniMax's documentation states that MiniMax-M3.1-Flash-Preview is available only through M Plan and MiniMax Code. It is called with an M Plan subscription key, and MiniMax published no per-token price for the preview.
How does its reasoning differ from MiniMax M3?
M3 offered thinking on or off. M3.1-Flash-Preview always reasons, and you tune how deeply with the effort parameter: low, medium, high, xhigh or max. MiniMax recommends max as the default.
What inputs and context length does it support?
Text, image and video input, with text output, and a context window of up to 1,000,000 tokens.