Chinese startup MiniMax announced on October 19, 2026 the launch of MiniMax M3.1 Flash, a language model focused on agent tasks, coding and computer use. The company confirmed that the anonymous Space Bunny model, which processed more than 200 trillion tokens, was an early version of MiniMax M3.1 Flash.
What happened?
Space Bunny operated in stealth on OpenRouter in late September, leading token usage rankings. MiniMax revealed the identity on the same day as the official launch of MiniMax M3.1 Flash. The model offers a 1 million token context window, pricing of $0.15 per million input tokens and $0.60 output, and promises about 70% reduction in agent task costs compared to the previous generation, with nearly three times faster inference speed.
Open weights for MiniMax M3.1 Flash are scheduled for October 23. The model is available via MiniMax API, M Plan and MiniMax Code.
Why it matters?
The launch of MiniMax M3.1 Flash reinforces competition among Chinese labs for efficient and cheap models for AI agents, following the wave started by DeepSeek. The stealth approach allows real-scale testing before formal announcement. Regardless of proprietary or third-party benchmarks, the focus on cost and speed attracts developers seeking alternatives to more expensive closed models.
Sources include MiniMax official announcement on X and reports from 36Kr and other outlets.
Image credit: MiniMax / Press
By GeekikiBot
