MiniMax Launches M3.1 Flash AI Model With 70% Lower Usage Costs

Chinese artificial intelligence company MiniMax launched its upgraded M3.1 Flash model on October 19, 2026, targeting developers and businesses that rely on AI to complete complex tasks. The company describes the release as a major improvement over its previous models, with enhanced capabilities in software development, computer use, cybersecurity and multi-step automation. Unlike conventional chatbots that primarily generate text in response to questions, M3.1 Flash is designed to help AI agents plan and execute tasks across different applications. According to MiniMax, the model can handle workflows involving several steps while maintaining performance at a lower cost, potentially making advanced AI automation more accessible to startups and businesses.

One of the most notable claims concerns the model’s speed and efficiency. MiniMax says M3.1 Flash processes online requests nearly three times faster than its predecessor and reduces usage costs for agent-based tasks by approximately 70%. The company also reports that the model can generate more than 200 tokens per second under favourable conditions, with median output speeds exceeding 170 tokens per second. These improvements are particularly relevant to AI agents, which may need to make numerous model calls while researching information, writing code, checking results and completing tasks. Faster processing can reduce waiting times, while lower costs can make it more practical to run agents continuously or deploy them across large numbers of users. These figures are based on MiniMax’s own disclosures and may vary depending on workload and deployment conditions.

M3.1 Flash also focuses heavily on software engineering and computer-use capabilities. MiniMax reports a score of 73.7% on the DeepSWE v1.1 coding benchmark, which evaluates an AI system’s ability to understand codebases, implement changes, fix bugs and validate its work. The company also reports a score of 61.7 on OSWorld V2, a benchmark designed to measure how effectively AI systems operate computers and complete tasks across applications. For businesses, these capabilities could support automated software maintenance, testing, data processing and repetitive administrative work. An AI agent could, for example, inspect a software project, identify a problem, modify the relevant code and run tests to check whether the change works. However, benchmark results do not guarantee that the model will perform equally well in every real-world environment, so independent testing remains important.

Cybersecurity is another area MiniMax has highlighted. The company reports an 82% score on CyberGym, a benchmark associated with discovering and validating software vulnerabilities. It also says its internal evaluations showed improvements in code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering. These capabilities could help security professionals identify weaknesses in software and investigate potential attacks more efficiently. At the same time, powerful cybersecurity capabilities can introduce risks if misused, particularly when AI systems are able to automate technical tasks at scale. Responsible deployment therefore requires appropriate access controls, monitoring and safeguards to ensure that security tools are used for authorised defensive work. MiniMax’s published results indicate its intended capabilities, but further independent evaluation would help establish how reliably the model performs on practical security tasks.

The model is also competitively priced. MiniMax lists API rates of $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. This pricing is intended to make M3.1 Flash attractive to developers building applications that process large volumes of information or require repeated AI interactions. The model is currently available through the MiniMax API, the company’s M Plan subscription and MiniMax Code. MiniMax has said that model weights will be released later, with October 23 announced as the planned date. An open-weight release could give developers greater flexibility to evaluate and deploy the model in their own environments, although the final availability and licensing terms will determine how it can be used.

The launch comes as Chinese AI developers compete with major American technology companies to deliver capable models at lower prices. MiniMax is attempting to differentiate itself by combining advanced agent capabilities with high processing speed and relatively inexpensive inference. If its claims hold up under broader testing, M3.1 Flash could appeal to companies developing coding assistants, business automation tools and cybersecurity products. Nevertheless, the model’s long-term impact will depend on its reliability, performance outside company-run benchmarks and the practical costs of running it at scale. For now, the release illustrates a growing trend in the AI industry: competition is no longer based solely on model intelligence, but also on how quickly, economically and safely AI systems can complete useful work.