MiniMax M3 / 채팅
Flagship multimodal foundation model supporting text, image, and video inputs with text output. Features a 1M-token context window via MiniMax Sparse Attention (MSA), cutting per-token compute to ~1/20 of previous gen at full context. Excels at long-horizon agentic work, coding, and tool use. Native multimodal training from step zero ensures deep semantic alignment. Scores 59.0% on SWE-Bench Pro and 83.5 on BrowseComp, surpassing Opus 4.7
Playground Chat
AI 모델과 대화를 시작하세요. 무엇이든 질문할 수 있습니다.
AI가 생성한 응답의 정확성은 다를 수 있습니다.
MiniMax M3Reliable Streaming Chat Completions
MiniMax M3 is a production chat model with streaming defaults, temperature and top-p sampling, stop sequences, and configurable max output tokens.
At a Glance
M3 Chat Capabilities
Straightforward chat completions for product surfaces.
Streaming by Default
stream defaults to true so tokens arrive as they are generated for chat UIs.
Sampling Controls
Temperature default 0.7 and top-p default 0.95 for tone and diversity control.
Stop Sequences
Bound generation so parsers and UIs never see trailing chatter.
Max Output Tokens
Cap completion length for cost control and predictable latency.
How It Works
Simple chat completions path.
POST Messages
Send system and user turns in chat completions schema.
Stream Tokens
Consume SSE for immediate UI paint.
Bound Output
Use stop sequences and max tokens for parser-safe replies.
Tune Sampling
Adjust temperature and top-p for brand voice.
M3 Product Domains
Where dependable chat is the product.
Customer Chat
Streaming support replies with stop-bounded answers.
Content Assist
Drafts, rewrites, and summaries.
Ops Automation
Ticket drafting and internal Q&A.
Search Answers
Grounded replies over retrieved context.
Prompt Tips
Keep M3 responses reliable.
Put role, constraints, and output format in the system message.
Keep stream on for chat surfaces; disable only for batch jobs.
Bound generation when you will parse JSON or field lists.
MiniMax M3 Quickstart
Streaming production chat.
curl -X POST "https://api.powertokens.ai/v1/chat/completions" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "MiniMax-M3",
"messages": [
{"role": "user", "content": "Draft a concise out-of-office reply for a client email."}
],
"stream": true,
"temperature": 0.7
}'Technical Specifications
Confirmed parameters and runtime execution protocols.
가격 상세
이 모델의 실제 요금은 API 요청에서 전달된 특정 매개변수를 기반으로 동적으로 계산됩니다. 아래는 구체적인 조합과 해당 가격입니다:
| 모달리티 | 입력 크레딧 | 출력 크레딧 | 입력 가격 | 출력 가격 | 암시적 캐시 적중 | 명시적 캐시 적중 | 캐시 생성 |
|---|---|---|---|---|---|---|---|
| 0 - 512K | 600/ 1M Tokens | 2,400/ 1M Tokens | $0.600 | $2.400 | $0.120 120/ 1M Tokens | $0.120 120/ 1M Tokens | -- |
| Over 512K | 1,200/ 1M Tokens | 4,800/ 1M Tokens | $1.200 | $4.800 | $0.240 240/ 1M Tokens | $0.240 240/ 1M Tokens | -- |
Models from the Same Channel
Explore complementary models and alternative versions from the same provider channel.
MiniMax-M3.1-Flash
MiniMax-M3.1-Flash
Next-gen efficiency-focused multimodal model (428B MoE) for agentic workflows, coding, and computer use . Features a 1M-token context window and tunable thinking depth (low–max) . Scores 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, and 82% on CyberGym, while delivering 200+ TPS output speed at 70% lower agent-task cost than its predecessor
MiniMax M2.7
MiniMax-M2.7
MiniMax-M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. Built to actively participate in its own evolution, M2.7 integrates advanced agentic capabilities through multi-agent collaboration, enabling it to plan, execute, and refine complex tasks across dynamic environments.
MiniMax M2.7 Highspeed
MiniMax-M2.7-highspeed
A high-speed variant of MiniMax’s flagship M2.7 LLM, delivering 100 TPS—~60% faster than standard M2.7—with identical top-tier quality. It features a 204,800-token context window, near-Opus-level SWE performance, and recursive self-improvement. Optimized for low-latency coding, office tasks, and agent workflows, it balances extreme speed, reliability, and cost-efficiency for high-throughput real-world applications。
MiniMax M2.5
MiniMax-M2.5
MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon the coding expertise of M2.1 to extend into general office work, reaching fluency in generating and operating Word, Excel, and Powerpoint files, context switching between diverse software environments, and working across different agent and human teams
Recommended Related Models
Explore complementary video and multimodal models with your unified API key.
GLM-5.3
glm-5.3
Zhipu AI's flagship text model optimized for complex software engineering and long-horizon Agent tasks. Features a 1M-token context window with mandatory reasoning (3 levels: low/high/max). Coding capability improved 50% over GLM-5.2 on Z.ai Code Bench; scores SOTA on Terminal Bench 3.0. Excels in cybersecurity tasks (cyber vulnerability discovery)
GLM-5.3 Flash
glm-5.3-flash
Z.ai's native multimodal model (320B-A18B) with text, image, video, and file input. Features hybrid sparse + linear attention, reducing attention compute 3x and KV cache 4.4x vs. GLM-5.3. Scores 57 on Artificial Analysis Intelligence Index—matching Claude Opus 4.8—with 1M context and 128K max output. Excels in agentic coding, vision-based Computer Use, and CAD reproduction
Qwen3.8 Flash
qwen3.8-flash
Alibaba's cost-efficient multimodal reasoning model. Supports text, image, and video inputs with text output. Features a native 1M-token context window for long documents, codebases, and agentic workflows. Excels in coding assistance, desktop interaction, chart analysis, and long-video understanding. Compatible with OpenAI/Anthropic protocols for seamless integration
GLM-5.2
glm-5.2
Flagship text model purpose-built for long-horizon agentic workflows. Features a 1M context window supporting project-level engineering in a single session. Excels at autonomous coding: can complete development, testing, and multi-platform deployment from a single prompt. Top open-weight model per Artificial Analysis; #1 globally on Code Arena. MIT-licensed and Day-0 optimized for domestic AI chips
Frequently Asked Questions
Everything you need to know before integrating this model.
Start Building with MiniMax M3 Today
Create an account in seconds to receive free test credits. No upfront payment or long-term contract required.