Reportedly: MiniMax releases M3.1 Flash: 3x faster inference, 70% lower cost
Reported by IT Home · not yet confirmed by the company or a second independent outlet. We update this page when it is.
MiniMax released the M3.1 Flash model, previously anonymous as Space Bunny, with nearly 3x faster inference and 70% lower cost. It scores 76.2 on BankerToolBench and 73.7% on DeepSWE v1.1. Availability outside China is not stated, IT Home reported.
China context
- Original name
- 稀宇科技
- Outside China
- Not stated in the sources yet
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China should verify API access and regional availability before planning integration, as the announcement did not specify global availability.
- For investors
- The price-performance improvement may intensify competition in the agentic AI market, but investors should seek independent validation of the claimed benchmarks and cost savings.
Translated from Chinese. Quotes and facts link to the original sources.
MiniMax announced the launch of its M3.1 Flash model, which had been operating anonymously as Space Bunny and processed over 200T tokens. The model delivers nearly 3x faster online inference and a 70% reduction in usage cost compared to its predecessor. It shows improvements in agentic tasks, coding, computer use, and cybersecurity, with reported scores of 76.2 on BankerToolBench and 73.7% on DeepSWE v1.1. The announcement did not specify availability outside China.
The reported 3x inference speedup and 70% cost reduction suggest architectural or serving optimizations, but no technical details were provided. The model's performance on DeepSWE v1.1 (73.7%) and BankerToolBench (76.2) indicates strong long-horizon task execution and financial compliance capabilities, though these are company-reported and not independently verified.
Developers using MiniMax's API can now access faster and cheaper inference for agentic and coding workloads, potentially reducing their operational costs. Competitors offering similar agentic models may face pressure to match these price-performance improvements.
The 70% cost reduction and 3x speed improvement make M3.1 Flash more attractive for high-volume, latency-sensitive applications such as autonomous agents and coding assistants, potentially lowering total cost of ownership for enterprises.
Independent verification of the benchmark scores and real-world latency/cost measurements would confirm the claimed improvements. A specific signal to check is whether third-party evaluations or customer case studies emerge that validate the 3x speedup and 70% cost reduction in production settings.