讯飞星辰MaaS

What is iFlytek Xingchen MaaS

iFlytek Xingchen MaaS is a one-stop large model fine-tuning and service platform launched by iFlytek, integrating model inference API, zero-code fine-tuning, and developer subscriptions. The platform aggregates mainstream large models such as DeepSeek, Qwen, GLM, Kimi, MiniMax, and iFlytek Spark, providing dual-mode model customization with one-sentence fine-tuning and professional fine-tuning. It offers developers the Astron Coding Plan coding subscription service, replacing pay-per-use billing with a fixed monthly fee, covering the entire chain of needs from model invocation and fine-tuning to AI programming.

Main features of iFlytek Xingchen MaaS

• Model inference: provides real-time inference services and batch inference capabilities, supporting high-concurrency API calls and one-click deployment.
• Model fine-tuning: supports zero-code fine-tuning, Notebook development, and model distillation, meeting the needs of different developers.
• Model hosting: provides own-model hosting and preset model hosting services, enabling efficient model deployment and management.
• Model evaluation: supports multi-dimensional effect evaluation, providing multiple scoring modes for fact-based and open-ended Q&A.
• Dataset management: provides preset datasets and custom dataset functions, supporting data upload and version management.
• Data cleaning and augmentation: generates more training samples through data augmentation tools, improving data quality and model performance.
• Q&A pair extraction: supports automatic segmentation and extraction of Q&A pairs that conform to platform specifications from text files or website links.
• Prompt engineering: provides preset Prompt templates and customization functions, helping developers optimize instruction design.
• Experience center: supports online comparative experience of models before and after fine-tuning, enabling quick verification of model performance.
• Large model application platform: provides a one-stop large model application development environment, lowering the barrier to AI application development.
• Model Square API service: provides inference APIs for popular models such asDeepSeek-V4-Pro、Spark-X2-Flash、Qwen3.6-35B-A3B、GLM-5.1、Spark X2, supporting text generation and multimodal understanding, with some models offering free inference quotas.
• Astron Coding Plan: a monthly subscription coding service for developers, integrating multiple model foundations and connecting through OpenAI/Anthropic dual protocols toCursor、Claude Code、OpenClawand other mainstream programming tools.
• One-sentence fine-tuning: aimed at users with zero foundation, providing preset scenarios such as role, classification, and creation. Enter a scenario description to quickly generate a dedicated model without preparing training data.
• Professional model fine-tuning: supports uploading custom datasets (json/jsonl/csv format, recommended ≥2000 entries), performing deep fine-tuning through a complete toolchain, and monitoring loss functions and accuracy in real time.

Core advantages of iFlytek Xingchen MaaS

• Model ecosystem aggregation: one platform integrates mainstream domestic and international large models, allowing users to switch foundations as needed without separately connecting to multiple vendors' APIs.
• Dual-mode fine-tuning threshold: from zero-threshold rapid verification with one-sentence fine-tuning to enterprise-level deep customization with professional fine-tuning, covering the needs of users at all stages.
• Data engineering closed loop: A unique question-answer pair extraction and data augmentation tool that automatically converts unstructured documents into training data, greatly reducing data preparation costs.
• Developer Subscription Innovation: Coding Plan replaces pay-as-you-go billing with a fixed monthly fee, completely eliminating the risk of overage arrears, and is compatible with the mainstream AI coding tool ecosystem.

How to Use iFlytek Spark MaaS

• Registration and Access: Visit the iFlytek Spark MaaS official website https://training.xfyun.cn/modelSquare , complete account registration and login, obtain an API key, and quickly integrate via the OpenAI-compatible interface.
• Select a Model: Browse 70+ preset models in the model marketplace, and choose a suitable text, image, or multimodal model according to your scenario.
• Data Preparation: Upload a custom dataset, or use a preset dataset, and optimize data quality through data cleaning, augmentation, and question-answer pair extraction.
• Model Fine-Tuning: Choose no-code fine-tuning or Notebook development, configure parameters and submit a training task, and view training progress and metrics in real time.
• Deployment and Invocation: After training is complete, publish the model service, integrate it via API calls or batch inference, and verify the results in the experience center.
• Application Building: Connect with open-source applications such as OpenClaw and OpenCode, or develop a dedicated AI Agent on the large model application platform.

Usage Examples of iFlytek Spark MaaS

Text Model

Introduce the famous mountains of China

Vision Model

Draw a realistic photo, Autumn leaves, clear lake water, fallen leaves, dead wood, bridges in the distance, blue sky, white clouds, forests, rocks, nature, tranquility, rich colors, contrast, slow door shots, smooth water surface, wide angle or medium focal length lens。拍摄一张 realistic photo（写实照片），画面包含：秋叶、清澈湖水、落叶、枯木、远景桥梁、蓝天白云、森林、岩石，展现自然的宁静之美。要求色彩丰富、对比强烈，采用慢门拍摄技法使水面呈现丝滑效果，使用广角或中焦镜头取景。

iFlytek Spark MaaS Product Pricing

• Pay-as-You-Go：
  • DeepSeek-V4-Pro： Input 12 yuan/million tokens, output 24 yuan/million tokens.
  • Spark-X2-Flash： Input 1 yuan/million tokens, output 2 yuan/million tokens.
  • Qwen3.6-35B-A3B： Input 1 yuan/million tokens, output 3.6 yuan/million tokens.
  • GLM-5.1： Input 8 yuan/million tokens, cache hit 1.6 yuan/million tokens, output 28 yuan/million tokens.
  • Spark X2： Input 3 yuan/million tokens, output 3 yuan/million tokens.
• Subscription Plans：
  • Worry-Free Edition: First purchase ¥3.9/month (thereafter ¥19/month), supports models such as Spark-X2-Flash, Qwen3.6-35B-A3B (default), Qwen3.5-35B-A3B, Qwen3-Coder-Next-FP8, GLM-4.7-Flash, etc., with unlimited requests.
  • Professional Edition: ¥39/month, supports models such as Spark X2, GLM-5 (default), GLM-5.1, MiniMax-M2.5, KIMI-K2.5, DeepSeek-V3.2, Spark-X2-Flash, Qwen3.6-35B-A3B, GLM-4.7-Flash, Qwen3.5-35B-A3B, Qwen3-Coder-Next-FP8, Qwen3.5-397B-A17B, etc., with a maximum of 1,200 requests every 5 hours, 9,000 per week, and 18,000 per month.
  • High-Efficiency Edition: ¥199/month, supports the same models as the Professional Edition, up to 6,000 requests every 5 hours, 45,000 per week, and 90,000 per month.

Comparison of iFlytek Spark MaaS with similar competing products

Application scenarios of iFlytek Spark MaaS

• Intelligent customer service: Based on domain data fine-tuned models, build professional customer service robots to achieve precise Q&A and multi-turn dialogue.
• Content creation: Use text generation models to assist with news writing, marketing copy, novel creation, etc., improving content production efficiency.
• Code development: Through code large models, provide functions such as intelligent completion, bug fixing, and code review to accelerate the software development process.
• Data analysis: Combine large model capabilities to achieve report generation, data insights, and visual interpretation, lowering the threshold for data analysis.
• Intelligent office: Connect with AI Agents such as OpenClaw to automatically handle daily tasks such as email categorization, schedule management, and document organization.