Chinese AI firm MiniMax releases M3.1 Flash model

By Splitfeed EditorialLast updated Oct 20, 2026, 08:20 UTC

On October 19, 2026, Chinese AI startup MiniMax released a new large language model called M3.1 Flash, positioning it as a cost-efficient option for developers working on coding and agentic applications. The company claims the model reduces the cost of AI agent tasks by 70 percent compared to prior approaches. MiniMax also reported that the model was trained on 200 trillion tokens, a figure tied to an internal teaser campaign the company called Space Bunny. The release was reported by Reuters and picked up by a range of outlets on the same day.

The announcement matters because it arrives in a competitive moment for large language model development, where cost reduction has become one of the primary battlegrounds between labs. A claimed 70 percent cost cut for agentic tasks, if verified, would make M3.1 Flash a meaningful option for enterprises and developers building multi-step AI workflows. The model also represents continued momentum from Chinese AI firms entering a space long dominated by US-based companies. What remains genuinely in dispute is whether the stated cost reductions hold up under independent benchmarking. MiniMax made the claim, but no third-party verification of the figure has been detailed in available sources, and the specific methodology behind the comparison is not yet public.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, had published reactions at the time of writing. Pro-AI voices would typically treat a 70 percent cost reduction as strong evidence that AI capabilities are becoming more accessible and economically viable for a broader range of use cases. Anti-AI commentators would likely raise questions about the opacity of the benchmark claims, potential labor and data sourcing practices behind a 200 trillion token training run, and the broader implications of geopolitical competition driving rapid AI scaling. Middle Ground observers would probably call for independent verification before drawing conclusions, while acknowledging that competitive pressure from Chinese firms does appear to be accelerating cost declines across the industry.

The argument will sharpen once independent developers test M3.1 Flash on standardized agentic benchmarks, or once MiniMax publishes fuller technical documentation around the model's training setup and the cost comparison methodology underlying the 70 percent figure.