Current outcome
MiniMax M3.1 Flash is fully released with open weights: 428B-parameter MoE, 70% lower agent-task cost, community license with revenue restrictions, and a 73.7% DeepSWE v1.1 score.
Progress timeline
2 material updates- #01
MiniMax正式发布M3.1 Flash：每百万Token最低0.15美元，23日开源
MiniMax officially released M3.1 Flash, confirming it was the anonymous 'Space Bunny' model, publishing benchmarks (62.2% Automation-Bench, 73.7% DeepSWE, 91.0% BabyVision), setting API pricing at $0.15 per million input tokens and $0.60 output, and planning to open the weights on October 23.
Source evidence: theblockbeats · jtg
- #02
MiniMax M3.1 Flash开源：4280亿参数，Agent任务成本降70%
MiniMax released the open weights for M3.1 Flash on October 23 as planned, listed on Hugging Face and ModelScope; disclosed a 428B-total-parameter MoE (~23B active) with sparse attention, quantized training and DSpark speculative decoding, 70% lower agent-task cost and ~3x generation speed, under the MiniMax Community License (written authorization required above $20M annual revenue).
State after update: MiniMax M3.1 Flash is fully released with open weights: 428B-parameter MoE, 70% lower agent-task cost, community license with revenue restrictions, and a 73.7% DeepSWE v1.1 score.
Source evidence: theblockbeats