What happened
MiniMax announced M3.1 Flash, a next-generation model focused on agentic tasks, coding, computer use and cybersecurity. The company claims the model is nearly three times faster online, reaches more than 200 tokens per second in everyday use, and costs 70% less for agentic tasks than its predecessor. It is available through the MiniMax API, M Plan and MiniMax Code. MiniMax said model weights will be released later.
MiniMax presents M3.1 Flash as an evolution of an earlier system it says operated anonymously under the name “Space Bunny” and processed more than 200 trillion tokens. The company says the new model improves on agentic tasks, coding, computer use and cybersecurity, and reaches the level of leading models on complex agentic tasks, software development and real-world computer-use evaluations. These are claims from the model developer, not independently established results.
The company reports output speeds above 200 tokens per second in some everyday use, with median speeds above 170 tokens per second. It lists flat pricing of $0.15 per 1 million input tokens, $0.60 per 1 million output tokens and $0.015 per 1 million cached input tokens. MiniMax says the model is now available through its API, M Plan and MiniMax Code. The source does not specify regional availability, plan limits, rate limits, service-level commitments or whether every listed feature is available through each channel. Model weights are promised later, without a release date or license.
Why it matters
The announcement adds another low-cost, high-speed model to the competitive market for AI systems that perform multi-step work, write software and interact with computers. If MiniMax’s claims hold in independent testing, the pricing could make repeated agentic workloads more affordable for developers and businesses. The source does not establish how the model performs outside the cited benchmarks, whether access is unrestricted, or when downloadable weights will appear.
The main practical significance is the combination of speed, task capability and stated cost reduction. Agentic applications often involve many model calls, so lower token prices could reduce operating costs for software development, computer-use workflows and other multi-step systems. Higher output speed could also make interactive coding and automation feel more responsive. The effect will depend on actual usage patterns, latency beyond token generation and the model’s reliability on long or error-prone tasks.
MiniMax cites Automation-Bench, DeepSWE and Agents’ Last Exam, saying M3.1 Flash performs on par with top models at low task cost. The source provides no scores, test configurations, baselines, evaluator information or independent replication. It also does not document safety performance, data handling, privacy terms or the model’s behavior in production. Those gaps limit how confidently organizations can compare the announcement with competing systems or assess deployment risk.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Independent evaluations of the claimed results, real-world reliability and cybersecurity performance will be important. Users should also watch for the terms and limits attached to API, M Plan and MiniMax Code access, as well as the eventual release date and license for model weights.
Independent results should clarify whether the claimed parity with leading models holds across agentic execution, coding and computer-use tasks, rather than only in MiniMax’s reported comparisons. Practical testing should also examine failure recovery, tool-use accuracy, prompt sensitivity, latency and cybersecurity safeguards.
Access conditions remain partly unknown. The source names the MiniMax API, M Plan and MiniMax Code but does not state whether these offerings are free, subscription-based or subject to separate quotas; only token pricing for the API or model usage is documented. The timing, access terms and licensing for the planned weight release are also unknown. MiniMax says a more capable Pro model will arrive later, but gives no date or specifications.