iFlytek Releases Spark X2.6, Agent OS, and Xingchen TaaS Platform
At the 2026 Global 1024 Developer Festival, iFlytek released the device-cloud integrated Spark X2.6, expanded the context window to one million Tokens, and open-sourced the Spark X2.6-4B on-device model, with device-cloud routing accuracy increasing from 81.2% to 96.1%. At the same time, it debuted the Agent OS agent collaboration platform and the Xingchen TaaS platform; the former is said to improve cross-system collaboration efficiency by 90% and enable more than 75% of processes to flow autonomously, while the latter builds a full-chain system from model capabilities to industry solutions. The iFlytek Open Platform also launched the XFYUN CLI development workbench, where a single command can invoke more than 700 AI atomic capabilities and more than 200 agent adaptation capabilities.
Why read this
AI score 60
iFlytek has released Spark X2.6, Agent OS, and Xingchen TaaS all at once, allowing readers to understand its industrial implementation path from models to agent collaboration.
Topics
Body · AI translation
Add crowdsourcing promotion copy Add crowdsourcing promotion copy end Add App-exclusive traffic guidance
On October 22, the 2026 iFlytek Global 1024 Developer Festival opened at the China National Convention Center in Beijing. Liu Qingfeng, Chairman of iFlytek, released a number of AI innovation achievements, including the device-cloud integrated Spark X2.6, the Xingchen TaaS platform, the Spark speech foundation model SparkAudio 1.0, and the Spark embodied multimodal interaction technology.
At the developer ecosystem conference held concurrently, the iFlytek Open Platform released a new-generation human-computer interaction solution for the first time, officially launching the agent collaboration platform Agent OS and a development workbench for the Agent ecosystem, driving AI from technical capabilities toward business implementation and making Tokens truly become productivity for thousands of industries.
Device-cloud integrated Spark X2.6 released, developer ecosystem flourishes
For AI to solve real-world scenario challenges, continuous evolution of model capabilities is indispensable. The Spark X2.6 released this time expands the context window to one million Tokens and simultaneously open-sources the Spark X2.6-4B on-device model. The first-ever release of the Spark large model device-cloud integrated collaboration architecture introduces a device-cloud routing collaborative training mechanism based on a critique model, raising device-cloud routing accuracy from 81.2% to 96.1%, further balancing on-device response speed with cloud understanding capability.
iFlytek Chairman Liu Qingfeng releases on site
The Spark speech foundation model SparkAudio 1.0 is also launched simultaneously. Trained on fully domestic computing power, the model upgrades capabilities around speech recognition, simultaneous speech interpretation, and real-time interaction. Through innovations such as a semi-supervised audio encoder, dynamic segmented end-to-end modeling, and dual-paradigm collaborative decoding, measured results show a 20% performance improvement, a 15% reduction in response latency, and a 60% drop in invocation cost, further enhancing machines' ability to understand sound.
Only with a prosperous ecosystem can there be a great future for artificial intelligence. Currently, the iFlytek Open Platform has reached 11.87 million developers, with the number of large-model enterprise developers growing 104% year on year. The 2026 iFlytek AI Developer Competition attracted 52,120 participating teams from 351 cities in 15 countries, setting a new record in scale. Since the open-sourcing of the Spark X2.5-4B on-device model, downloads have reached 2.56 million, ranking among the top four on-device models globally. Model capabilities, development tools, and the open ecosystem continue to work in concert, driving AI from simple Q&A toward complex task processing and enabling more Tokens to create business value.
Anthropomorphic, universal, easy to use: new-generation human-computer interaction solution released
Targeting scenarios such as homes, robots, and public spaces, the iFlytek Open Platform launched a new-generation human-computer interaction solution for the first time, advancing smart hardware interaction upgrades around the three directions of anthropomorphism, universality, and ease of use.
Liu Kewei, Chief Technology Officer of the iFlytek Open Platform, presents the developer ecosystem release
In natural interaction, an end-to-end model integrated with microphone arrays strengthens spatial acoustic perception, helping devices locate sound sources and identify conversation partners; combined with the device-cloud collaboration architecture, it enables instant on-device response and deep cloud understanding. The WallEX smart space solution, through open IoT protocols and the Harness platform, supports multi-brand, multi-device linkage, with local control response as fast as 700 milliseconds.
In complex environment adaptation, the iFlytek Robot Super Brain platform integrates panoramic cameras and circular microphone arrays, supporting interaction among up to 5 people simultaneously, and still achieves 90.1% speech recognition accuracy under -10dB strong noise conditions. The AIUI multilingual interaction platform covers 8 major regions including Europe, the Middle East, and Southeast Asia, supporting seamless communication in more than 40 languages. Meanwhile, the platform provides dedicated fine-tuned models for scenarios such as children's companionship, elderly care, and teaching practice, among which the children's dedicated interaction large model achieves 95% recognition accuracy.
Yin Jing, 1024 Product Experience Officer of iFlytek, demonstrates multilingual interaction on site
To lower the development threshold, the platform also offers software and hardware products such as voice interaction boards, multimodal interaction boards, Spark Super Brain Board 2.0, and embodied interaction modules, and accelerates products from prototype validation to mass production through reference designs and complete-device solutions.
Agent OS debuts, enabling enterprise agents to form a "relay race"
Addressing the needs of complex business division of labor, cross-system collaboration, and process automation, the iFlytek Open Platform released Agent OS for the first time and, together with APA, built an enterprise-level Agent capability system. Among them, Agent OS is responsible for organizational collaboration and task scheduling, while APA is responsible for automated execution of business processes, driving AI from model invocation toward an end-to-end business closed loop.
Agent OS organizes collaboration among employees, agents, and business systems through "scenario packages," and optimizes skills and processes by combining organization-level context engineering and continuous evolution mechanisms. Practical data shows that cross-system collaboration efficiency improves by 90%, and more than 75% of processes can flow autonomously.
Zhou Hanyue, GuideX Product Manager of iFlytek, demonstrates APA on site
APA supports business personnel in encapsulating daily operations such as form filling, inventory checking, and initiating approvals into reusable skills through screen-recording operations or natural language descriptions, which are then invoked on demand by agents to connect business systems such as ERP and CRM.
In industry practices, at the China Energy Investment Corporation, iFlytek's intelligent procurement review accuracy reaches 97%; in overseas practices with e-commerce KA customers, the "Super Performance Optimizer" drives a 40% increase in the per-capita scale of accounts managed and a 50% increase in ROAS.
Invoke 700+ AI capabilities with a single command, open platform takes the lead in Agentization
For Agent development, the iFlytek Open Platform has launched the XFYUN CLI development workbench. With a single command, developers can invoke more than 700 AI atomic capabilities and more than 200 agent adaptation capabilities, simplifying capability onboarding, authentication, and environment configuration processes.
Through the XFYUN CLI development workbench, 700+ AI capabilities can be invoked with a single command
AIUI Code supports developers in describing requirements, with an Agent invoking plugins to assist development; the Interactive R&D Digital Employee supports updating robot configurations through voice commands, enabling installation and configuration, instant updates, and scenario adaptation.
Live demonstration of the Interactive R&D Digital Employee
GuideX has upgraded its integrated software-hardware offering, packaging real-world needs from "fragmented procurement" into composable, deliverable, and sustainably operable integrated solutions. Among them, AStudio, as a new-generation all-scenario AI productivity workbench, handles requirement intake. GuideX, as an interactive intelligence platform integrating Agents, virtual humans, and multimodality, undertakes the entire subsequent process starting from the configuration and assembly stage. In the live demonstration, an airport concierge customized for the airport scenario saw GuideX and AStudio work together to complete the configuration and adaptation of the interactive application. In addition, the platform builds reusable Agent resources around Skills, MCP, connectors, CLI, and industry knowledge, driving AI application development toward standardized, engineered, and large-scale delivery.
Spark TaaS debuts, making Tokens real productivity
As large model applications accelerate toward deployment, how to truly convert Token consumption into business value has become key to industry development. The newly debuted Spark TaaS platform builds a full-chain system from model capabilities to industry solutions, opening a new stage in the co-evolution of Tokens and Agents.
At the foundation, it relies on the Spark MaaS platform, supporting the Spark large model and a variety of open-source large models; the middle layer uses model scheduling, agent engines, knowledge engineering, tool orchestration, and runtime governance to make Tokens serve business needs more precisely and efficiently; the application layer covers scenarios such as simultaneous interpretation and translation, digital employees, office efficiency, knowledge Q&A, and intelligent marketing, and provides products and solutions for industries including enterprise services, finance, healthcare, education, and energy.
This means developers no longer need to build complex technical systems from scratch, and enterprises no longer need to remain at the stage of purchasing Tokens and calling models; instead, they can obtain application capabilities oriented toward real business. Tokens are integrated into real business processes as they are consumed, so that every Token invocation can advance a business task and complete a task, and Solutions can truly play a role in high-value scenarios across thousands of industries. Let Tokens not spin idle, and let value be visible.
From model breakthroughs and interaction upgrades to agent collaboration and full-chain development and delivery, the iFlytek Open Platform is continuously connecting models, tools, agents, and industry knowledge, working with developers to bring AI innovation results into thousands of industries. In the future, the iFlytek Open Platform will continue to work with the broad community of developers to jointly build an open and prosperous AI industry ecosystem, enabling technological innovation to create broader social value. Statement: This is an exclusive article from Sina.com; reproduction without authorization is prohibited. tech_web_article_v2015_block_bottom Non-targeted 300*250 button 17/09 wenjing begin
Non-targeted 300*250 button end