iFLYTEK Spark X2.6: a 293B MoE trained entirely on domestic Chinese computing
iFLYTEK launched Spark X2.6 at its Global 1024 Developer Festival on Oct 22, 2026: a 293B-parameter MoE (30B active) with a 1M-token context, which the company says is the first model to reach a step change in coding and agent capability while trained entirely on domestic computing hardware, alongside an open-source X2.6-4B on-device model.
Why it matters
The training story is the product: dynamic parallel training that cuts long-sequence training on domestic chips from 4.5 months to 1.2 months, plus a 4.5x inference efficiency gain — a direct answer to hardware restrictions.
What you could build with it
A startup serving Chinese enterprise customers could build a long-document coding assistant on Spark X2.6's 1M-token context and domestic-hardware inference story, offering data-residency-friendly deployment that foreign-hosted models cannot match.
Does it hold up?
Trial access is available through iFLYTEK's Spark platform and the 4B model is open-sourced, but benchmark and efficiency claims are the company's own and independent verification is not yet available.
Built with iFLYTEK Spark X2.6
- iFlytek launches Spark X2.6, claiming training entirely on domestic computingSuperpower Daily · Launch analysis of the domestic-computing claim, the 293B-A30B specs, and the aggressive trial pricing (0.9 yuan per million input tokens).
- iFLYTEK Launches X2.6 End-Cloud AI Model Powered Entirely by Domestic Computing ResourcesKuCoin News · Festival report covering the cloud model, the open-source X2.6-4B on-device model, and the three new inference modes.
- iFLYTEK released Spark X2.6 at the 1024 Developer Festival on 10-22Shanghai Metals Market · Flash report with the Feixing-2 cluster numbers: 10,000 cards, training time cut from 4.5 to 1.2 months, KV-cache hit rate up from 63% to 95.7%.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.