MiniMax
Released October 19, 2026

MiniMax-M3.1-Flash

MiniMax-M3.1-Flash by MiniMax is a multimodal coding model with 1M context, always-on reasoning, five effort levels, and text, image and video input.

Inputs
Text
Image
Video
Outputs
Text

Model Overview

Capabilities, design details, and architectural traits

MiniMax-M3.1-Flash - Frontier Multimodal Coding Model with Tunable Thinking Depth

MiniMax-M3.1-Flash is a MiniMax preview model for frequent coding, financial research, multimodal understanding and frontend work. Its defining design choice is always-on reasoning: the model thinks before every answer and there is no supported way to turn thinking off. Instead, you tune how deeply it thinks through an effort setting.

TraitDetail
ReasoningAlways enabled; sending thinking disabled or effort none returns HTTP 400 with the message requires adaptive thinking
Effort levelsLow, Medium, High, Xhigh and Max; Max is the default recommendation
Context window1,000,000 tokens
Inputs and outputText, image and video inputs; text output, with a native visual encoder
ArchitectureMixture-of-experts with 428B total parameters, about 23B active, and sparse attention
Reasoning outputSeparate reasoning_content field; integrations must handle absent values
AccessSubscription plans and MiniMax Code only; no open weights published

A reasoning model that cannot be silenced

Unlike MiniMax-M3, where thinking can be enabled or disabled per request, M3.1-Flash removes the off switch entirely. Lowering effort is the documented way to reduce thinking latency and tokens, not disabling it. For a model carrying the Flash name, defaulting to Max effort is a deliberate design opinion: depth of reasoning is treated as a dial, never a switch.

Built for everyday expert work

MiniMax positions the model around real, verifiable tasks inside MiniMax Code: diagnosing and fixing performance issues with browser profiling and flame charts, building financial workbooks that separate historical data from assumptions, and turning reference photos into interactive Three.js reconstructions. The emphasis is on evidence from tests and measurements rather than plausible code, with prompt caching supported for API integrations.

Benchmark Performance

Independent evaluations · Artificial Analysis

29.1%
Intelligence

Accuracy & Capability Details

Humanity's Last Exam48.0%
SciCode - Scientific Coding53.4%
Long Context Reasoning79.7%

Compare Models Side-by-Side

Evaluate specifications, pricing, and independent benchmark indices

Model Details
General Info
ProviderMiniMaxAnthropicOpenAI
Release DateOctober 19, 2026November 9, 2026November 17, 2026
Knowledge Cutoff---
Context & Limits
Context Window1M1M
1.1M
Best Context Window
Pricing (per 1M tokens)
Input Pricing
$0.15
Best Input Pricing
$8$8
Output Pricing
$0.60
Best Output Pricing
$40$40
Modalities
Inputs
textimagevideo
textimagefile
textimagefile
Outputs
text
text
text
Benchmarks (0-100)
Intelligence Index29.1
64.6
Best Intelligence Index
62.1
Coding Index---
Agentic Index---

Benchmark comparison not available

Select at least two models that have available benchmark evaluation data to render the performance comparison charts.

Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.

Explore more from MiniMax

Other models by MiniMax

Top AI Models

Leading alternatives by intelligence score

View all