Home
Xunfei Xinghuo X2-Flash Launches with Enhanced Domestic Computing Power and 256K Long-Text Support
On April 29, iFLYTEK officially launched the new Spark X2-Flash model and opened its API interface, marking a new efficiency milestone for large model applications built on the domestic computing power ecosystem.
The model uses the current mainstream MoE (Mixture of Experts) architecture with a total of 30 billion parameters. Its standout feature is support for an ultra-long context of up to 256K. Notably, the Spark X2-Flash was fully trained on Huawei's Ascend 910B cluster, showcasing the synergy of domestic software and hardware for deep learning training.

In core performance, the Spark X2-Flash delivers significant gains in agent and code generation capabilities. Third-party tests show its performance on complex tasks—such as processing in-depth research reports, managing and invoking skills, and executing system control—has reached the level of top-tier industry models with trillions of parameters.
On cost, which developers care about, the Spark X2-Flash excels. In identical workflow tests, its token consumption is just one-third that of current mainstream large models, greatly lowering the barrier to building complex agent applications. For instance, when creating intricate video generation skills, the model not only quickly grasps the requirements but also offers detailed explanations covering skill structure to core functions.

Technically, the Spark X2-Flash is the first to combine DSA (Sparse Attention) and MTP (Multi-Token Prediction) on domestic chips. This innovation resolves the slow long-text training issue on domestic computing platforms, boosting training efficiency by 4.5 times compared to similar-scale clusters. Additionally, for agent reinforcement learning scenarios, the model improves sampling inference efficiency by over two times through combined algorithmic and engineering optimizations, effectively easing performance bottlenecks in long interaction scenarios.
Currently, applications like AstronClaw and Loomy have completed integration first. The model also deeply aligns with international mainstream agent frameworks such as OpenClaw and Claude Code, offering global developers a more cost-effective domestic computing power option.
Related article
Suno to Watermark Songs Amid Legal Battles
Suno, the platform enabling users to generate AI-created music, has unveiled new features to label platform-produced tracks, restrict downloads, and update community standards to curb unauthorized replicas. These updates arrive as Suno confronts mult
Musk Admits Grok Build Leaked User Code, Promises to Erase All Historical Data
Elon Musk directly addressed the privacy controversy surrounding Grok Build, beginning with a simple "True" to confirm the incident's validity. He pledged that all user data previously uploaded to SpaceXAI would be permanently erased, stating, "not a
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Related Special Topic Recommendations
Comments (0)
0/500
On April 29, iFLYTEK officially launched the new Spark X2-Flash model and opened its API interface, marking a new efficiency milestone for large model applications built on the domestic computing power ecosystem.
The model uses the current mainstream MoE (Mixture of Experts) architecture with a total of 30 billion parameters. Its standout feature is support for an ultra-long context of up to 256K. Notably, the Spark X2-Flash was fully trained on Huawei's Ascend 910B cluster, showcasing the synergy of domestic software and hardware for deep learning training.

In core performance, the Spark X2-Flash delivers significant gains in agent and code generation capabilities. Third-party tests show its performance on complex tasks—such as processing in-depth research reports, managing and invoking skills, and executing system control—has reached the level of top-tier industry models with trillions of parameters.
On cost, which developers care about, the Spark X2-Flash excels. In identical workflow tests, its token consumption is just one-third that of current mainstream large models, greatly lowering the barrier to building complex agent applications. For instance, when creating intricate video generation skills, the model not only quickly grasps the requirements but also offers detailed explanations covering skill structure to core functions.

Technically, the Spark X2-Flash is the first to combine DSA (Sparse Attention) and MTP (Multi-Token Prediction) on domestic chips. This innovation resolves the slow long-text training issue on domestic computing platforms, boosting training efficiency by 4.5 times compared to similar-scale clusters. Additionally, for agent reinforcement learning scenarios, the model improves sampling inference efficiency by over two times through combined algorithmic and engineering optimizations, effectively easing performance bottlenecks in long interaction scenarios.
Currently, applications like AstronClaw and Loomy have completed integration first. The model also deeply aligns with international mainstream agent frameworks such as OpenClaw and Claude Code, offering global developers a more cost-effective domestic computing power option.
Suno to Watermark Songs Amid Legal Battles
Suno, the platform enabling users to generate AI-created music, has unveiled new features to label platform-produced tracks, restrict downloads, and update community standards to curb unauthorized replicas. These updates arrive as Suno confronts mult
Musk Admits Grok Build Leaked User Code, Promises to Erase All Historical Data
Elon Musk directly addressed the privacy controversy surrounding Grok Build, beginning with a simple "True" to confirm the incident's validity. He pledged that all user data previously uploaded to SpaceXAI would be permanently erased, stating, "not a
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation











