MiniMax Cuts AI Agent Costs By 70 Percent

MiniMax's new M3.1 Flash model cuts costs for complex agentic tasks by 70% and offers three times faster online inference speeds.

MiniMax has introduced a new AI model, M3.1 Flash, designed to significantly lower the cost and increase the speed of building complex AI applications. According to a report from The New Stack, the company claims this new model can reduce the overall cost of "agentic tasks" by as much as 70% while delivering online inference speeds nearly three times faster than its predecessor, the M3 model. To achieve this, MiniMax has set an aggressive price point for M3.1 Flash at just $0.15 per million input tokens and $0.60 per million output tokens. This pricing is half of the discounted rate for the more powerful M3 model, which charges $0.30 and $1.20 for input and output tokens, respectively, on requests up to 512,000 tokens. The launch positions M3.1 Flash as a highly competitive option for developers focused on building scalable and cost-effective AI-powered services.

The key to understanding the impact of M3.1 Flash lies in its focus on "agentic tasks." This term refers to complex, multi-step workflows where an AI model acts more like an autonomous agent than a simple chatbot. Instead of just answering a single question, an AI agent might plan a series of actions, use external tools like APIs or web browsers, and self-correct its approach to achieve a goal, such as booking a multi-city trip or conducting detailed market research. These operations are notoriously expensive because they often require numerous sequential calls to the AI model, with each step consuming tokens and adding to the final bill. A 70% cost reduction specifically for these workloads suggests that M3.1 Flash is not just a smaller, cheaper model but has likely been architected for efficiency in reasoning and tool-use loops. The threefold increase in inference speed also directly addresses the latency problem inherent in agentic systems, enabling more responsive, real-time interactions that are crucial for user-facing applications.

The release of M3.1 Flash is a clear reflection of a major trend shaping the generative AI landscape: model tiering and specialization. Leading AI labs like OpenAI, Anthropic, and Google have all adopted a strategy of offering a spectrum of models, from powerful but expensive flagship versions to smaller, faster, and much cheaper alternatives. This allows developers to choose the right tool for the job, optimizing for the specific balance of intelligence, speed, and cost that their application requires. For many tasks, the absolute best model is often overkill, and a "good enough" model that is an order of magnitude cheaper can unlock commercial viability. MiniMax's move shows it is competing aggressively in this tiered market. By specifically targeting agentic tasks, the company is also betting on what many believe is the next frontier for AI—moving beyond simple text generation to creating autonomous systems that can perform meaningful work. This specialization could give MiniMax an edge in a rapidly growing segment of the AI application market.

For founders, CTOs, and development teams, the arrival of M3.1 Flash presents a compelling opportunity to re-evaluate their AI strategy. Teams currently using more expensive models for agent-like workflows should seriously consider benchmarking M3.1 Flash to see if it meets their performance needs at a fraction of the cost. The potential for a 70% cost reduction could fundamentally change the economics of an AI product, enabling new pricing models or making previously unprofitable features viable. However, it is crucial to approach these claims with a degree of caution. The performance metrics are, for now, self-reported by MiniMax. The next step for the developer community will be to conduct independent, real-world tests to verify not only the speed and cost savings but also the qualitative performance of the model. The key question will be how much reasoning and instruction-following capability was sacrificed to achieve this price point, and whether it remains powerful enough for truly complex agentic tasks. The market will be watching closely to see if M3.1 Flash delivers on its promise and how competitors respond.

For developers and CTOs, the release of M3.1 Flash signals a significant drop in the operational cost of building AI agents. This makes sophisticated, multi-step AI workflows more economically viable, potentially unlocking new applications that were previously too expensive to run at scale.

The dramatic price reduction of MiniMax M3.1 Flash intensifies competition in the AI model market. Companies can now build more powerful AI agents for less, lowering the barrier to entry for startups and forcing established players to reconsider their own pricing strategies.