ChinaiFlytek cut one training run from 4.5 months to 1.2.
It says Spark X2.6 ran entirely on China-made chips and prices input below DeepSeek-V4.1-Flash.
iFlytek says its 10,000-card Feixing-2 cluster lifted cache hit rates from 63% to 95.7% and inference efficiency 4.5 times.
On SWE Verified, iFlytek’s own results put it behind DeepSeek-V4.1-Flash.
Chairman Liu Qingfeng says 60% of AI tasks will run locally.
How each outlet framed it
- Zhidx leans favorable
- emphasizes the domestic-chip training claim (Feixing II, 4.5 to 1.2 months) and the 60/30/10 local-cloud task split
Sources: Zhidx