iFlytek's iFly Open Platform has announced a significant upgrade to the iFly Spark API platform, which includes the Spark Max and 4.0 Ultra models. These models have seen a 70% increase in generation speed, along with notable breakthroughs in key performance areas such as logical reasoning, creative capabilities, and internet search.

The upgraded Spark Max and 4.0 Ultra models demonstrate faster speeds and higher accuracy in logical reasoning, especially in handling complex tasks, providing detailed reasoning processes. In content creation, the new models can generate articles with logical and practical value, such as Xiaohongshu copywriting, and have improved the accuracy rate of Function Call instructions to over 90%.

WeChat Screenshot_20240925074332.png

Additionally, iFly Spark has introduced the new Spark Max-32K sequence model, which supports over 40,000 Chinese characters input, effectively enhancing the ability to handle long text tasks and meeting the needs of industries such as law and finance. This upgrade maintains the same price while increasing performance, achieving "more for the same price" to lower the barriers for developers.

WeChat Screenshot_20240925074249.png

iFly Spark also offers a wealth of resources including beginner guides, development documentation, and a community forum. The Spark Debug Center supports rapid debugging and parameter adjustments, helping developers quickly build code for specific application scenarios. In terms of pricing, iFly Spark has introduced a permanently free Lite version and offers a buy-one-get-one deal for the 4.0 Ultra's first order, with the Max version available for free with 100 million Tokens, significantly reducing developers' trial costs.

WeChat Screenshot_20240925074304.png

In terms of performance, iFly Spark V4.0 has undergone a comprehensive upgrade in text generation, language understanding, knowledge questioning, logical reasoning, and mathematical abilities, benchmarking against the internationally leading GPT-4 Turbo, and achieving superiority in several core capabilities.

Spark Large Model API Matrix: https://xinghuo.xfyun.cn/sparkapi