This week, Alibaba's newly open-sourced CozyVoice 3 supports nine languages and achieved SOTA performance across multiple established benchmarks. AgentScope releases upgrades to better support production-grade AI agent development and deployment. Meanwhile, the Teams plan was launched to provide enterprises with scalable, secure, and production-ready AI-powered development.
Alibaba open sources CozyVoice 3 with zero multilingualism‑Shot speech model that advances cutting-edge speech synthesis
Alibaba has open sourced CosyVoice 3, a multilingual speech synthesis model that significantly outperforms previous products in terms of content consistency, speaker similarity, and prosody naturalness.
The core of the model is A novel voice tokenizer Powers a wide range of speech understanding and analysis capabilities, including automatic speech recognition (ASR), speech emotion recognition, language identification, speech event detection, and speaker analysis. Furthermore, in this model, New differentiable reward optimization (DiffRO) method Directly optimize speech tokens after training. This technique is particularly useful for large-scale language model (LLM)-based speech synthesis models, allowing for more accurate and expressive speech production.
trained on a vast dataset of 1 million hours of audiothe model has two variations—Featuring 500 million and 1.5 billion parametersrespectively. Supported by CozyVoice 3 9 languages Includes English, Chinese, German, Spanish, French, Italian, Japanese, Korean, and Russian. Achieve state-of-the-art (SOTA) performance across multiple established benchmarks.
As a zero-shot speech generation model, CosyVoice 3 represents a major advance in multilingual, high-fidelity speech synthesis, paving the way for more versatile, natural, and context-aware voice applications in real-world scenarios such as virtual assistants, audiobook/podcast narration, customer service automation, short video and game narration production, and generation of interactive dialogue in educational settings.
Developers around the world can now access models on Hugging Face, Github, ModelScope, and experience them on our demo site.
CosyVoice3 delivers superior performance in content consistency
AgentScope releases major upgrade to production-ready AI agents
Announced by Alibaba Doyi Research Institute Upgrade to AgentScope, an AI agent development frameworkfocused on improved production readiness to better support a broader range of application scenarios, a more powerful infrastructure, and the development and deployment of production-grade AI agents.
On the application side, AgentScope has evolved from providing basic functionality such as deep exploration, browser usage, and planning to providing agent applications optimized for out-of-the-box scenarios. for example, alias is a production-ready agent that dynamically switches between general and specialized modes (financial analysis, data science, etc.), supports one-click local deployment, and serves as a customizable template for domain-specific solutions. evo traders Simulate a real-world investment team and refine your strategy using the ReMe memory system. Both Alias and EvoTraders offer highly customizable interfaces for configuring toolsets, prompts, and orchestration logic that can be used out of the box or extended to integrate your own tools, data, and business workflows.
At its foundational layer, AgentScope introduces Agent Skills, a plug-and-play dynamic skills framework that allows agents to flexibly configure functionality for complex tasks. It also introduces AgentScope-Studio, a visual development environment with dual-view message streams, ReAct state tracing, and OpenTelemetry integration to improve debugging efficiency and developer experience.
In terms of deployment, AgentScope-Runtime v1.0 introduces a “white box” paradigm that balances ease of use and flexibility. Developers now have precise control over the lifecycle of their agent applications without sacrificing simplicity. The runtime natively supports multi-agent collaboration, enables customizable sharing of sessions, memory, and toolkits for efficient coordination, and also enables cross-framework orchestration to ensure consistency from development to production. It also features a multi-type sandbox matrix, covering local and cloud environments across a variety of device types, providing secure and isolated execution of browser control, file operations, and mobile automation, all with MCP-based extensibility.
The AgentScope ecosystem also includes SparkChat with a built-in web UI for instant visual interaction, and AgentScope-Java v1.0 for enterprise Java stacks.
The latest AgentScope upgrades are published on GitHub, giving enterprises and developers access to a more robust, production-ready agent development ecosystem.
Qoder launches Teams plan to accelerate enterprise-grade AI coding
Qoder, an agent coding platform built for real-world software development, has launched Qoder Teams. This is a new offer designed to help businesses of all sizes with scalable, secure, and production-ready AI-powered development.
Teams plans offer enterprise-grade features such as advanced knowledge services, enhanced context awareness, and specification-driven development. We also offer a flexible suite of tools, including Qoder IDE, Qoder Plugin for JetBrains IDE, and Qoder CLI, to seamlessly integrate into diverse team workflows and existing development environments.
Key enterprise features further enhance the service, including centralized billing, single sign-on (SSO) integration, and future support for shared credit pools. This upcoming feature will allow organizations to purchase AI credits in bulk and allocate them dynamically across teams, enabling flexible usage-based resource management.
By integrating AI capabilities into the development environment, Qoder helps engineering teams streamline coding, reduce context switching, and accelerate delivery without compromising security or code quality.
