Chinese AI developer MiniMax Group Inc. launches M3.1 Flash, an update to its M3 large language model first released in June. The company positions the new model as faster and cheaper, specifically targeting the cost of running AI agents that typically require many model calls to complete a single task.
MiniMax says M3.1 Flash reduces AI agent costs by 70%. Separate coverage also notes that the company halves its per-token prices alongside the release. However, both outlets imply the impact on end-user “agent bills” may vary depending on how frequently an application calls the model and how pricing translates from token usage to total operational costs. The reporting frames the launch as a cost-focused iteration rather than a major change in functionality.
