iFLYTEK released Spark X2.6, a device-cloud model family, at its 2026 Global 1024 Developer Festival on October 22. The company says it is the first model in China to reach a step change in coding and agent capability while being trained entirely on domestic computing power. Chairman Liu Qingfeng said the model was trained in the past two to three months.
The cloud model uses a 293B-A30B mixture-of-experts design, meaning 293 billion total parameters with about 30 billion active per token, and supports a context window of 1 million tokens. On iFLYTEK's own benchmarks, it placed first on SWE Pro and SWE Multilingual and second on SWE Verified, and ranked in the top group on enterprise operations agent tasks and the MCP-Atlas tool-use test. Limited-time pricing is 0.9 yuan per million input tokens, 0.08 yuan per million cached input tokens and 3.6 yuan per million output tokens.
Much of the presentation focused on training and inference on domestic chips. iFLYTEK says dynamic parallel training for long-sequence tasks works around the memory bandwidth limits of domestic accelerators. On its Feixing-2 cluster, training on 2,000 billion tokens of 256K-length data took 1.2 months instead of 4.5 months. For inference, the company described distributed inference across heterogeneous domestic hardware and a four-tier KV Cache architecture built entirely on domestic hardware.
The company also open-sourced Spark X2.6-4B, an on-device model aimed at robots, cars and AI PCs. iFLYTEK says it beats its predecessor Spark X2.5-4B as well as Qwen3.5-4B, Gemma4-E4B and Qwen3.5-9B on search, document work, scheduling, spreadsheets and multi-step tool calls. The previous 4B model, open-sourced on September 1, is now among the four most downloaded on-device models worldwide, according to the company. A router decides whether each task runs on the device, in the cloud or across both, and local data is masked before anything is sent to the cloud.
iFLYTEK also debuted SparkAudio 1.0, a speech foundation model trained on domestic compute that combines speech recognition, translation and semantic understanding with real-time speaker separation, sound-scene understanding and emotion recognition. It is already running in iFLYTEK Input Method 16.0, iFLYTEK Hearing, its smart notebooks and its translators.
All benchmark and efficiency figures are iFLYTEK's own. The cloud and on-device models are available to try on the Spark website.
SEE ALSO: Zhipu AI's GLM-5.3-Flash Runs on 100,000 Domestic Chips · BAAI's FlagOS 2.2 Expands Its Cross-Chip Operator Libraries