MiniMax M3.1 Flash claims three‑fold faster inference and a 70% reduction in usage cost.
MINIMAX M3.1 FLASH LAUNCHES WITH 70% COST CUT AND 3X SPEED
*MiniMax rolls out M3.1 Flash, a next‑gen LLM that claims three‑fold faster inference and a 70% drop in usage cost. Benchmarks place it alongside top models in coding, automation and cybersecurity.*
MiniMax announced today the launch of M3.1 Flash, a flash‑grade large language model built on the anonymous Space Bunny experiment that processed more than 200 trillion tokens. The company claims the new model delivers a generational leap in agentic tasks, coding, computer use and cybersecurity while slashing online inference time to nearly three times faster than its predecessor. Cost is cut by 70%, positioning M3.1 Flash as the most economical option for high‑volume AI workloads. The rollout arrives as enterprises scramble for AI that can handle complex, multi‑step processes without exploding budgets.
What benchmarks prove M3.1 Flash’s edge?
MiniMax M3.1 Flash ranks among the top on AutomationBench‑AA, handling real enterprise SaaS workflows end‑to‑end. It scores 76.2 on BankerToolBench, excelling at multi‑step planning, document consistency, and financial compliance. In coding, it hits 73.7% on DeepSWE v1.1, matching leading models on complex software tasks that require independent code understanding, feature development, bug fixing and testing. For computer use, it posts 61.7 points on OSWorld V2 and 29.6 on Agents’ Last Exam, leading all Chinese models. Cybersecurity performance peaks at 82% on CyberGym, with internal tests confirming strength in code audit, vulnerability patching, penetration testing, malware analysis and reverse engineering.
Speed and cost: the numbers behind the claim
M3.1 Flash delivers online inference nearly three times faster than its predecessor. In everyday use, token throughput exceeds 200 TPS, with median speeds consistently above 170 TPS. Cost efficiency follows: usage cost drops 70%, with flat pricing per million tokens at $0.15 for input, $0.60 for output and $0.015 for cached input. The model’s aggressive sparsity drives these gains without sacrificing benchmark performance. MiniMax positions the model as the most cost‑effective option for agentic tasks, setting a new standard for high‑performance, low‑cost AI.
How MiniMax plans to distribute the model
M3.1 Flash is available through the MiniMax API, the subscription‑based M Plan, and the MiniMax Code desktop client. Model weights will be released later, allowing developers to run the model locally. The company also teases a forthcoming “Pro” tier that will push intelligence boundaries further. Access points include platform.minimax.io for API keys and agent.minimax.io for code downloads. MiniMax frames the rollout as part of its broader mission to democratize intelligence while maintaining a competitive edge.
Implications for cybersecurity and enterprise automation
The 82% CyberGym score signals a model capable of low‑cost, high‑quality security work, from vulnerability discovery to reverse engineering. Enterprises can embed M3.1 Flash in security pipelines to automate code audits and penetration testing, reducing reliance on expensive human analysts. AutomationBench performance suggests the model can orchestrate multi‑tool SaaS workflows across sales, marketing, finance and HR, delivering end‑to‑end solutions. Combined speed, cost and capability make M3.1 Flash a potential disruptor for firms seeking AI‑driven efficiency without inflating budgets.
Frequently asked questions
What performance gains does MiniMax M3.1 Flash claim over its predecessor?
MiniMax says M3.1 Flash offers nearly three times faster online inference and reduces usage costs by 70% compared with the prior model.
How does M3.1 Flash perform on cybersecurity benchmarks?
The model scores 82% on CyberGym and internal tests show strong results in code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering.
What are the pricing rates for M3.1 Flash?
Pricing is $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens.
If MiniMax’s performance claims hold up in real‑world deployments, M3.1 Flash could reset the economics of AI‑driven automation and security. Faster, cheaper, and benchmark‑competitive, the model forces rivals to confront a new cost baseline. The coming release of model weights and the promised Pro tier will test whether MiniMax can sustain its advantage or simply ride a wave of hype.
Sources: Hacker News (MiniMax M3.1 Flash blog)