iFlytek to Release Spark X2.5 Base Model on Sept. 7

Summarized by NextFin AI

iFlytek will formally release its Spark X2.5 base model with 293 billion parameters on Sept. 7, upgrading coding and agent capabilities.
On Sept. 1, the company plans to open-source Spark X2.5-4B and Spark X2.5-1.7B edge models supporting up to 1 million token context windows.
The edge models target improved performance in agent, math and general understanding for vehicle, smart-hardware and IoT scenarios.
At its 2026 interim results briefing, iFlytek signaled a new flagship model trained on domestic computing power at the Global 1024 Developer Festival.

NextFin News — iFlytek said it will formally release its Spark X2.5 (293 billion parameters) base model on Sept. 7, further upgrading capabilities in coding and agents.

On Sept. 1 the company plans to open-source two edge-side general models, Spark X2.5-4B and Spark X2.5-1.7B. Both natively support a context window of up to 1 million tokens and focus on improving agent, mathematics and general-understanding performance for vehicle, smart-hardware and Internet-of-Things scenarios.

At its 2026 interim results briefing, iFlytek had indicated it would launch a new flagship general large model trained entirely on domestic computing power at this year’s Global 1024 Developer Festival.

Explore more exclusive insights at nextfin.ai.

Insights

What is the Spark series model evolution path?

What does a 293 billion parameter base model mean?

How does a 1 million token context window work technically?

What are the use cases for edge-side general models?

How does iFlytek compete in the current Chinese AI model market?

What is the demand for AI models in vehicle and IoT scenarios?

How are open-source edge models impacting the industry?

What new capabilities does Spark X2.5 bring compared to X2?

Why is iFlytek open-sourcing models on Sept. 1 before the base release?

What was announced at the 2026 interim results briefing regarding domestic computing?

How will the Global 1024 Developer Festival shape iFlytek strategy?

What is the future of fully domestic computing power training for AI?

How might Spark X2.5 influence smart hardware development?

What challenges exist in training models entirely on domestic computing power?

How does parameter size affect deployment on edge devices?

Can open-source models maintain competitiveness against closed counterparts?

How does Spark X2.5 compare to other 200B parameter models?

Which competitors also offer 1 million token context windows?

How does iFlytek open-source strategy compare to other Chinese tech firms?

What historical cases show success in edge-side AI deployment?

NextFin.Al

No Noise, only Signal.

Open App