智能体
www.aibase.com ·
讯飞星火X2.5发布：MoE架构加持，代码与智能体能力升级
讯飞发布星火X2.5大模型，采用MoE架构，参数规模293B-A30B，重点提升代码生成和智能体开发能力，并基于国产算力完成全流程训练与推理。同时开源4B和1.7B边缘模型，支持超长上下文和本地部署。
AI推荐理由 讯飞星火X2.5发布，采用MoE架构，参数规模293B-A30B，强调代码与智能体能力提升，并基于国产算力完成训练推理，同时开源4B和1.7B边缘模型。信息具体，但来源为AIbase转载，非官方首发，且缺乏独立验证，故质量分中等偏高。对AI应用和国产算力生态有实际价值，兴趣分较高。
70
正文
讯飞星火X2.5发布：MoE架构带来显著提升，代码编写与智能体能力实现跨越式升级
AIbase基地
发布于AI新闻 · 3分钟阅读 · 2026年9月7日
10
近日，科大讯飞正式发布了新一代星火X2.5大模型。作为科大讯飞技术进步的又一重要里程碑，该模型在架构设计、核心能力以及国产算力适配等方面实现了全面升级与突破。
在底层技术架构与参数规模方面，星火X2.5采用了先进的MoE（混合专家）架构，参数规模达到293B-A30B。凭借这一架构优势，模型在代码编写和智能体（Agent）应用开发能力上进行了深度优化与专项增强，同时持续强化其在数学逻辑和综合语义理解方面的核心能力。
尤为值得一提的是，星火X2.5实现了全自控算力基础设施。它基于国产算力平台完成了从端到端训练到推理的全流程，并持续攻克和优化了训练与推理中与长序列相关的技术难题。目前，该模型已在科大讯飞开放平台正式上线，向开发者及企业用户全面开放。
除了大规模主模型外，科大讯飞还开源了两款轻量级端侧模型：星火X2.5-4B和1.7B。这些端侧模型不仅支持高达100万Token的超长上下文处理，还可本地部署，完美匹配长文档深度处理、代码辅助开发、复杂数据分析等多种应用场景。
星火X2.5MoE架构国产算力代码编写
本文来自AIbase日报
扫码查看
欢迎来到【AI日报】栏目！这是您每日探索人工智能世界的指南。每天，我们为您呈现AI领域的热点话题，聚焦开发者，帮助您了解技术趋势，并了解创新的AI产品应用。
——由AIbase日报团队创作
© 版权所有 AIbase基地 2024，点击查看来源 -
查看原文（中文正文为 AI 翻译）
Xunfei Xinghuo X2.5 Launch: MoE Architecture Provides Significant Enhancement, Code Writing and Intelligent Agent Capabilities Achieve Leapfrog Upgrades
AIbase基地
Published inAI News · 3 min read · Sep 7, 2026
10
Recently, iFLYTEK officially released the new Xinghuo X2.5 large model. As another important milestone in iFLYTEK's technological advancement, this model has achieved comprehensive upgrades and breakthroughs in architecture design, core capabilities, and domestic computing power compatibility.
In terms of underlying technical architecture and parameter scale, Xinghuo X2.5 adopts an advanced MoE (Mixture of Experts) architecture, with a parameter scale of 293B-A30B. Leveraging this architectural advantage, the model has made in-depth improvements and specialized enhancements in code writing and intelligent agent (Agent) application development capabilities, while continuously strengthening its core capabilities in mathematical logic and comprehensive semantic understanding.
Notably, Xinghuo X2.5 has achieved full self-controlled computing infrastructure. It has completed all processes from end-to-end training to inference based on China's domestic computing power platform, and has continuously overcome and optimized technical challenges related to long sequences in training and inference. Currently, the model is officially launched on iFLYTEK's open platform, offering full access to developers and enterprise users.
In addition to the main model with a large scale, iFLYTEK also opened source two lightweight edge-side models: Xinghuo X2.5-4B and 1.7B. These edge-side models not only support ultra-long context processing up to 1 million Tokens, but can also be deployed locally, perfectly matching diverse application scenarios such as deep processing of long documents, code assistance development, and complex data analysis.
Xinghuo X2.5MoE ArchitectureDomestic Computing PowerCode Writing
This article is from AIbase Daily
Scan to view
Welcome to the [AI Daily] column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with hot topics in the AI field, focusing on developers, helping you understand technical trends, and learning about innovative AI product applications.
—— Created by the AIbase Daily Team
© Copyright AIbase Base 2024, Click to View Source -