ChainThink news, September 3: According to a recent announcement by fal, the AI inference platform fal has launched its Turbo version just one week after the release of MiniMax H3 Max, reducing generation latency by approximately 2.5 times and cutting prices in half.

The Turbo version is based on post-training of MiniMax H3 and optimized specifically for fal's inference system. In official tests with 5-second, 768p video generation, latency dropped from 3.49 seconds on H3 Max to 1.40 seconds.

In terms of quality, fal's internal testing shows the Turbo version achieves 97% of the original H3 Max’s performance, with claims that prompt understanding, aesthetic quality, and overall output remain largely intact—still surpassing both the original MiniMax H3 and other benchmarked models.

Regarding pricing, the standard rate for Turbo is $0.04 per second, half the standard rate of H3 Max ($0.08 per second). During the launch promotion, the rate drops to $0.01 per second, costing approximately $0.60 to generate a one-minute video.