MiniMax M3.1 Flash API
MiniMax M3.1 Flash is an open-weight multimodal model with a 1M-token context window, adjustable thinking effort, and agentic task execution across SaaS systems.
MiniMax M3.1 Flash is MiniMax's next-generation Flash model, belonging to a series built for high-frequency, high-intensity, large-scale agent calls. It accepts text, image and video input natively and produces text output, with a context window of 1M tokens. Thinking is always on and adjustable across five effort levels—low, medium, high, ultra-high and max—with max as the default. A sparse attention architecture reduces the compute cost of response generation, and the model is available through OpenAI-compatible and Anthropic-compatible API endpoints with tool calling support.
The model is designed for agentic work. It can execute tasks across SaaS systems, plan across multiple steps with cross-document consistency, and handle long agentic coding sessions including repository generation, refactoring and terminal tasks. It also targets computer use in real environments, cybersecurity work such as code auditing, vulnerability patching, penetration testing and malware analysis, plus frequent coding and research tasks spanning financial research, multimodal understanding and frontend development.
In everyday use, output speed exceeds 200 tokens per second, with median output speed above 170 tokens per second. Online inference is about 3x faster than the previous generation, while overall usage cost on agentic tasks drops to roughly 30% of the previous generation. An early anonymous version, Space Bunny, processed over 200 trillion tokens. Independent testing has flagged limitations: Artificial Analysis reported an intelligence index of about 42 to 43 and a high overconfident wrong-answer rate, and one user review says benchmark results do not hold up in real-world comparison. Fact-checking is advised where factual accuracy matters.
Updated:
MODEL ID
minimax-m3.1-flash- Context
- 1.049M
- Max output
- 524.288K
- Recommended API
- messages
- Input
- text / image / video
- Output
- text
Highlights
- 1M-token context window
- Native text, image and video input
- Five adjustable thinking effort levels
- Sparse attention for lower compute cost
- OpenAI- and Anthropic-compatible APIs
- Open weights released October 23, 2026
Providers
Compare API prices across providers. Select a provider to view all charges and conditions.
| $0.15 | $0.6 | $0.015 | |
| $0.15 | $0.6 | $0.015 |
What $10 buys
- Input
- 66.7M tokens
- Output
- 16.7M tokens
Estimated from current input and output prices, before discounts.
Benchmarks
- 76.2
BankerToolBench
points
- 73.7%
DeepSWE v1.1
percentage
- 61.7
OSWorld V2
points
- 29.6
Agents' Last Exam (ALE)
points
- 82%
CyberGym
percentage
- 56.2%
SWE-Pro
percentage
- 57.0%
Terminal Bench 2
percentage
- 1495
GDPval-AA
Elo
Scores as published by the model provider.
Reasoning
- Default effort
- max
- Always enabled
- Yes
- Available effort levels
- max / xhigh / high / medium / low
Quick Start
Drop-in code to call this model with Felo API Platform compatible APIs. Start in the Playground when you want to test prompts before writing code.
Get your API key
Create a key from your Felo API Platform account and set it as an environment variable.
export FELO_API_KEY="YOUR_API_KEY"Make your first request
Use this model ID with the Felo API Platform endpoint. Optional request headers such as X-Request-Id can help troubleshooting.
minimax-m3.1-flashThis model is reasoning-capable. If your selected protocol and model route expose reasoning controls, preserve the returned reasoning metadata when continuing a conversation.
const response = await fetch("https://openapi.felo.ai/api/v1/messages", {
method: 'POST',
headers: {
Authorization: `Bearer ${process.env.FELO_API_KEY}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
"model": "minimax-m3.1-flash",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}),
});
if (!response.ok) throw new Error(await response.text());
console.log(await response.json());Using third-party SDKs
Use compatible SDKs by setting baseURL to https://openapi.felo.ai/api/v1. Use the Anthropic-compatible base URL for Claude-style clients. https://openapi.felo.ai/api.
Enable streaming
Add stream: true to receive responses as server-sent events.
curl -N https://openapi.felo.ai/api/v1/messages \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-m3.1-flash",
"max_tokens": 1024,
"messages": [
{
"role": "user",
"content": "Hello"
}
],
"stream": true
}'Endpoint
Felo API Platform exposes compatible model-calling surfaces. Pick the surface your app or agent client already supports.
Messages
Anthropic-compatible messages for Claude-style clients.
POST
https://openapi.felo.ai/api/v1/messagesParameters
modelstringminimax-m3.1-flashPublic model ID to call.max_tokensinteger—Maximum generated tokens.messagesarray—Conversation messages for Chat Completions.seed———stop———min_p———toolsarray—Tool calling definitions following compatible schema.top_k———top_p———logprobs———reasoning———logit_bias———temperaturenumber—Controls response variety when supported.tool_choicestring or object—Controls whether a tool should be called.top_logprobs———response_formatobject—Request structured output when supported.presence_penalty———reasoning_effort———frequency_penalty———include_reasoning———repetition_penalty———structured_outputs———Frequently asked questions
What is MiniMax M3.1 Flash?
MiniMax M3.1 Flash is MiniMax's next-generation Flash model, belonging to a series built for high-frequency, high-intensity, large-scale agent calls. It accepts text, image and video input natively and produces text output, with a context window of 1M tokens. Thinking is always on and adjustable across five effort levels—low, medium, high, ultra-high and max—with max as the default. A sparse attention architecture reduces the compute cost of response generation, and the model is available through OpenAI-compatible and Anthropic-compatible API endpoints with tool calling support. The model is designed for agentic work. It can execute tasks across SaaS systems, plan across multiple steps with cross-document consistency, and handle long agentic coding sessions including repository generation, refactoring and terminal tasks. It also targets computer use in real environments, cybersecurity work such as code auditing, vulnerability patching, penetration testing and malware analysis, plus frequent coding and research tasks spanning financial research, multimodal understanding and frontend development. In everyday use, output speed exceeds 200 tokens per second, with median output speed above 170 tokens per second. Online inference is about 3x faster than the previous generation, while overall usage cost on agentic tasks drops to roughly 30% of the previous generation. An early anonymous version, Space Bunny, processed over 200 trillion tokens. Independent testing has flagged limitations: Artificial Analysis reported an intelligence index of about 42 to 43 and a high overconfident wrong-answer rate, and one user review says benchmark results do not hold up in real-world comparison. Fact-checking is advised where factual accuracy matters.
How is usage billed?
Your bill depends on the provider you use, its prices and your usage. Longer inputs or other conditions may have different prices.
What input modalities does MiniMax M3.1 Flash support?
It accepts text, image and video input natively and returns text output.
How large is its context window?
The model supports a context window of 1M tokens.
Is thinking mode configurable?
Yes. Thinking is always on and adjustable across five effort levels: low, medium, high, ultra-high and max, with max as the default.
What tasks is it suited for?
It targets multi-step business workflow automation across SaaS tools, long agentic coding sessions, computer use, cybersecurity tasks and frequent coding and research work.
How does it compare with the previous Flash generation?
Online inference speed is about 3x that of the previous generation, and overall usage cost on agentic tasks is reduced to about 30% of the previous generation.
What should users watch out for?
Artificial Analysis reported an intelligence index of about 42 to 43 and a high overconfident wrong-answer rate. Fact-checking is advised for use cases that require factual accuracy.