Ant Group's Bai Ling Big Model Launches Ling-2.6-flash: Efficient Optimization + Free API Trial

Ant Group's Bai Ling Big Model recently launched the Instruct model Ling-2.6-flash with 104B total parameters and 7.4B active parameters. Previously, it was tested anonymously as Elephant Alpha on OpenRouter, where it saw continuous growth in calls, topping the Trending list for days with daily token usage of 100B level.

Ling-2.6-flash optimizes for inference efficiency, token efficiency, and Agent scenarios: Its hybrid linear architecture boosts inference speed to up to 340 tokens/s on 4 H20 cards; token efficiency calibration reduces consumption to 1/10 of similar models; Agent scenario enhancements lead to excellent performance in multiple evaluations.

To facilitate developer experience, Bai Ling Big Model offers a one-week free API trial on OpenRouter and its official platform. After the free period, it will be charged by usage, with the official platform retaining a daily free quota of 500,000 tokens. Additionally, BF16, FP8 versions of the model will be open-sourced soon.