MiniMax M3.1 Flash: 70% Cheaper, 3x Faster Inference
MiniMax released M3.1 Flash, an agentic AI model that delivers competitive performance across automation, software development, computer interaction, and security tasks while reducing computational cost by 70% compared to its predecessor. Benchmark results show 73.7% on complex code analysis and development tasks, 82% on vulnerability discovery and penetration testing, 76.2 on multi-step business process execution, and strong scores on real-world computer interaction benchmarks. The model achieves output throughput of 200+ tokens per second, with typical (P50) performance at 170+ TPS.
The pricing structure—$0.15 per million input tokens, $0.60 per output, and $0.015 for cached inputs—targets cost-sensitive agent workloads where high token volume and latency matter. M3.1 Flash is positioned as the efficiency tier in MiniMax’s lineup, with a more advanced Pro model coming later. The release establishes a new baseline for cost-effective agentic deployment, particularly for enterprise automation, security operations, and developer-facing tools.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.