MiniMax: MiniMax M1 (extended)

Text input Text output Unavailable
Author's Description

MiniMax-M1 is a large-scale, open-weight reasoning model designed for extended context and high-efficiency inference. It leverages a hybrid Mixture-of-Experts (MoE) architecture paired with a custom "lightning attention" mechanism, allowing it to process long sequences—up to 1 million tokens—while maintaining competitive FLOP efficiency. With 456 billion total parameters and 45.9B active per token, this variant is optimized for complex, multi-step reasoning tasks. Trained via a custom reinforcement learning pipeline (CISPO), M1 excels in long-context understanding, software engineering, agentic tool use, and mathematical reasoning. Benchmarks show strong performance across FullStackBench, SWE-bench, MATH, GPQA, and TAU-Bench, often outperforming other open models like DeepSeek R1 and Qwen3-235B.

Key Specifications
Cost
$$$$
Context
128K
Released
Jun 17, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Stop Presence Penalty Logit Bias Temperature Seed Structured Outputs Frequency Penalty Max Tokens Include Reasoning Tool Choice Top P Min P Tools Reasoning
Features

This model supports the following features:

Tools Reasoning Structured Outputs
Performance Summary

MiniMax-M1 demonstrates exceptional speed and cost efficiency, consistently ranking among the fastest models and offering highly competitive pricing across multiple benchmarks. This aligns with its design for high-efficiency inference and its "lightning attention" mechanism. However, the model's performance on the provided baseline benchmarks is highly inconsistent. While it achieved a strong 99.0% accuracy in Email Classification, indicating proficiency in specific classification tasks, its accuracy was 0.0% across Ethics, Reasoning, and General Knowledge benchmarks. This suggests significant limitations in handling complex ethical dilemmas, multi-step reasoning problems, and broad factual recall in these specific test sets. The long duration for Email Classification (2354404ms) also stands out as an anomaly compared to other tasks, despite its high accuracy. The model's reliability cannot be assessed from the provided data as no reliability ranking was given. Overall, MiniMax-M1 appears to excel in specific, well-defined tasks like email classification, but exhibits critical weaknesses in more open-ended or complex reasoning and knowledge-based domains within these baseline tests.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0
Completion $0

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Novita
Novita | minimax/minimax-m1:extended 128K $0 / 1M tokens $0 / 1M tokens
Chutes
Chutes | minimax/minimax-m1:extended 512K $0 / 1M tokens $0 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration
Other Models by minimax