On October 22, the 2026 iFLYTEK Global 1024 Developer Festival officially kicked off at the China National Convention Center in Beijing. Liu Qingfeng, Chairman of iFLYTEK, unveiled a suite of breakthrough AI innovations on-site, including the device-cloud integrated iFLYTEK Spark X2.6, the Xingchen TaaS Platform, the SparkAudio 1.0 speech foundation model, and Spark Embodied Multimodal Interaction Technology.
At the concurrent Developer Ecosystem Conference, the iFLYTEK Open Platform introduced its next-generation human-machine interaction solution alongside the official launch of Agent OS—an agent collaboration platform—and an agent-centric development workbench. These releases aim to drive AI beyond isolated technical capabilities toward real-world business implementation, transforming raw tokens into tangible productivity across diverse industries.
Launch of Device-Cloud Integrated Spark X2.6 as the Developer Ecosystem Thrives
Solving real-world challenges with AI requires continuous evolution in model capabilities. The newly released Spark X2.6 expands its context window to the million-token level, accompanied by the open-sourcing of the on-device Spark X2.6-4B model. Making its debut, the Spark device-cloud collaborative architecture introduces an evaluator-model-based joint training mechanism for device-cloud routing. This leap improves routing accuracy from 81.2% to 96.1%, striking a more seamless balance between millisecond-level on-device responsiveness and deep cloud-based understanding.

iFLYTEK Chairman Liu Qingfeng delivering the keynote launch
Concurrently, the SparkAudio 1.0 speech foundation model was launched. Trained entirely on domestic computing infrastructure, the model delivers across-the-board upgrades in speech recognition, simultaneous interpretation, and real-time interaction. Powered by innovations such as semi-supervised audio encoders, dynamic segmentation end-to-end modeling, and dual-paradigm collaborative decoding, benchmark tests demonstrate a 20% boost in overall performance, a 15% reduction in latency, and a 60% cut in inference costs—greatly advancing machine understanding of auditory signals.
A thriving ecosystem is the bedrock of AI’s long-term future. Today, the iFLYTEK Open Platform connects 11.87 million developers, with enterprise developers utilizing large models growing 104% year-over-year. The 2026 iFLYTEK AI Developer Competition drew a record-breaking 52,120 teams from 351 cities across 15 countries. Since its open-source release, the Spark X2.5-4B on-device model has surpassed 2.56 million downloads, ranking among the top four on-device models globally. The sustained synergy among model capabilities, developer tools, and an open ecosystem is shifting AI from simple Q&A to complex task execution, empowering more tokens to unlock substantive business value.
Anthropomorphic, Universal, and Accessible: Next-Generation Human-Machine Interaction
Tailored for diverse settings such as homes, robotics, and public spaces, the iFLYTEK Open Platform introduced its next-generation human-machine interaction solution. Built around three pillars—anthropomorphic, universal, and accessible—it comprehensively elevates the interactive experience of smart hardware.

Liu Kewei, CTO of iFLYTEK Open Platform, presenting the developer ecosystem updates
In natural interaction, the solution integrates microphone array end-to-end modeling to bolster spatial acoustic perception. This enables precise sound source localization and speaker identification. Backed by the device-cloud architecture, it pairs millisecond on-device response times with comprehensive cloud comprehension. Meanwhile, the WallEX Smart Space solution leverages open IoT protocols and the Harness platform to support seamless cross-brand, cross-device interoperability, driving local control response latencies down to as low as 700 milliseconds.
In terms of complex environment adaptation, the iFLYTEK Robot Super Brain Platform fuses panoramic cameras with circular microphone arrays, enabling simultaneous interactions with up to five people. Even under intense -10dB background noise, its speech recognition accuracy sustains a high 90.1%. The AIUI Multilingual Interaction Platform covers eight major global regions, including Europe, the Middle East, and Southeast Asia, supporting seamless conversations across more than 40 languages without manual switching. In addition, the platform offers dedicated fine-tuned models for niche domains such as child companionship, eldercare, and tutoring. Among these, the child interaction model achieves an accuracy rate of 95%.

Yin Jing, Product Experience Officer for iFLYTEK 1024, demonstrating multilingual interaction on stage
To substantially lower the barrier to development, the platform delivers a comprehensive suite of hardware and software kits, including speech interaction boards, multimodal interaction boards, the Spark Super Brain Board 2.0, and embodied interaction modules. Through standardized reference designs and turnkey solutions, it accelerates the transition of hardware products from prototyping to volume production.

Debut of Agent OS: Orchestrating Enterprise Agents into a Seamless Relay
To address complex enterprise needs regarding division of labor, cross-system collaboration, and workflow automation, the iFLYTEK Open Platform launched Agent OS and partnered with APA to construct an enterprise-grade agent framework. In this framework, Agent OS manages organizational coordination and global task scheduling, while APA handles the automated execution of specific workflows. Together, this dual-engine approach transforms AI from isolated model calls into end-to-end, fully closed-loop business operations.
Agent OS utilizes “Scenario Packs” to orchestrate collaboration between employees, intelligent agents, and business systems. Combined with organizational-level context engineering and self-evolving mechanisms, it continually refines skill matching and process flows. Real-world operational data shows that the system boosts cross-system collaborative efficiency by 90%, with over 75% of workflows achieving fully autonomous execution.

Zhou Hanyue, GuideX Product Manager at iFLYTEK, demonstrating APA live
APA enables business operators to package routine tasks—such as form filling, inventory queries, and approval routings—into reusable skills with a single click, using simple screen recording or natural language descriptions. These skills can then be invoked on demand by agents, seamlessly bridging disparate business systems like ERP and CRM.
In practical enterprise deployments, this system has demonstrated significant impact: at China Energy Investment Corporation (CHN Energy), iFLYTEK’s intelligent procurement evaluation achieved an accuracy rate of 97%; in overseas operations for leading e-commerce clients, the “Super Performance Optimizer” drove a 40% increase in ad spend managed per operator while improving ROAS by 50%.
Invoking 700+ AI Capabilities with a Single Command: Pioneering the Agent Era
For agent developers, the iFLYTEK Open Platform introduced the XFYUN CLI development workbench. With a single command, developers can invoke more than 700 atomic AI capabilities and over 200 agent-adapted features, radically streamlining interface integration, authentication, and environment setup.

Calling 700+ AI capabilities with a single command via the XFYUN CLI development workbench
AIUI Code allows developers to describe requirements in natural language while agents autonomously orchestrate plugins to assist with coding. Meanwhile, the Interactive R&D Digital Employee enables direct updates to robot configurations via voice commands, supporting instant configuration, hot-reloading, and scenario-specific tuning.

Live demonstration of the Interactive R&D Digital Employee
GuideX received a full software-hardware integrated upgrade, packaging fragmented physical-world demands into composable, deliverable, and sustainably manageable turnkey solutions. Within this ecosystem, AStudio serves as the next-generation, all-scenario AI productivity workbench handling front-end requirement inputs, while GuideX—an interactive intelligence platform integrating agents, virtual humans, and multimodal tech—leads the full-lifecycle deployment starting from assembly and configuration. In a live showcase of a customized “Airport Concierge,” GuideX and AStudio collaborated to configure and deploy a tailored interactive application in minutes. Furthermore, the platform has aggregated a rich set of reusable agent assets around Skills, MCP, Connectors, CLI, and vertical domain knowledge, driving AI application development toward standardization, engineering maturity, and scalable delivery.
Xingchen TaaS Debut: Turning Tokens into Tangible Productivity
As large model applications penetrate deeper into operational workflows, converting massive token consumption into genuine business and commercial value has become a pivotal challenge for the industry. The debut of the Xingchen TaaS (Task as a Service) Platform bridges the entire pipeline from core model capabilities to vertical industry solutions, ushering in a new era of co-evolution between tokens and agents.
Architecturally, Xingchen TaaS relies on the underlying Xingchen MaaS platform, providing comprehensive support for the iFLYTEK Spark model as well as mainstream open-source alternatives. The middle layer integrates intelligent model dispatching, agent engines, knowledge engineering, tool orchestration, and operational governance, ensuring that token flows accurately and efficiently align with business demands. The application layer spans generic scenarios—such as simultaneous interpretation, digital employees, office productivity, knowledge Q&A, and smart marketing—while delivering turnkey, end-to-end solutions for sectors including enterprise services, finance, healthcare, education, and energy.

The rollout of this platform means developers no longer need to construct complex underlying tech stacks from scratch, nor do enterprises need to remain stuck in an unrefined phase of simply purchasing tokens and calling raw APIs. Instead, organizations can directly obtain production-ready capabilities that generate business impact. As tokens are consumed, real value is infused straight into operational workflows—ensuring every call advances a business objective and completes a task. By establishing deep roots in high-value industrial workflows, the platform ensures that tokens never run idle and their value remains clearly measurable.
From base model breakthroughs and upgraded interaction paradigms to agent orchestration and end-to-end engineering delivery, the iFLYTEK Open Platform continues to break down the silos between models, tools, intelligent agents, and industry knowledge. Moving forward, the platform will continue collaborating with developers worldwide to cultivate an open, collaborative, and prosperous AI industrial ecosystem, transforming technological innovation into far-reaching social and economic value.