讯飞星辰MaaS

Overview

iFlytek Spark MaaS is an intelligent model fine-tuning and inference service platform launched by iFlytek, providing developers with full-chain engineering capabilities covering "data-model-service." The platform integrates modules such as a model marketplace, data processing, model training, performance evaluation, and service deployment, helping teams complete large model customization, hosting, and launch with lower barriers.

It supports multimodal capabilities such as text, images, and speech, providing features including zero-code fine-tuning, Notebook development, model distillation, and one-click deployment. It is also compatible with mainstream open-source model ecosystems and OpenAI-compatible interfaces, making it suitable for scenarios such as enterprise knowledge base Q&A, AI Agents, intelligent customer service, content generation, and AI programming.

Key Features

• Model Marketplace: Provides abundant built-in model resources covering common capabilities such as text generation, image understanding, and code programming.
• Model Inference: Supports real-time inference and batch inference, suitable for API integration and business system integration.
• Model Fine-Tuning: Supports efficient parameter optimization methods such as zero-code fine-tuning, Notebook development, and LoRA to meet different development depth requirements.
• Model Hosting: Supports hosting for built-in models and third-party models, facilitating unified management and deployment.
• Dataset Management: Supports uploading custom datasets, and platform built-in datasets can also be used, with version management available.
• Data Cleaning and Enhancement: Provides data enhancement and processing tools to improve training sample quality and model performance.
• Q&A Pair Extraction: Can automatically extract standardized Q&A data from text files or webpage links.
• Model Evaluation: Supports multi-dimensional performance evaluation, making it convenient to compare model performance before and after fine-tuning.
• Prompt Engineering: Provides preset templates and customization capabilities to help optimize instruction design.
• Deployment and Application Development: Supports one-click deployment and can be used to build applications such as AI Agents.
• Open Interface: Provides OpenAI-compatible interfaces for convenient integration with existing open-source tools and development environments.

Product Pricing

iFlytek Spark MaaS adopts a usage-based billing model, mainly charging based on the number of Tokens generated by model calls, and unit prices vary across different models. Some price examples mentioned in public information include:

• DeepSeek-V3.2: input about 2 yuan/million Tokens, output about 3 yuan/million Tokens
• DeepSeek-R1: input about 4 yuan/million Tokens, output about 16 yuan/million Tokens
• Kimi-K2 series: input about 4 yuan/million Tokens, output about 16 yuan/million Tokens
• Spark Max: input about 8 yuan/million Tokens, output about 40 yuan/million Tokens
• Qwen3 series: prices vary with model size, starting as low as about 0.3 yuan/million Tokens

Actual prices may change due to promotions, model versions, or billing policy adjustments. It is recommended to refer to the latest explanation on the official website.

Frequently Asked Questions

Which users is iFlytek Spark MaaS suitable for?

It is suitable for developers, enterprise technical teams, and product teams that need to call, fine-tune, or deploy large models and want to quickly build AI applications.

Does it support access to open-source models?

Yes. The platform is compatible with mainstream open-source model ecosystems and supports third-party model hosting.

Can it be used with low-code or no-code?

Yes. The platform provides zero-code fine-tuning capabilities and also supports more flexible Notebook-based development methods.

How can it be integrated into existing applications?

It can be integrated through the OpenAI-compatible interfaces provided by the platform, making it easier to connect with existing AI development tools and business systems.