Skip to main content
ScitiX Model Inference gives developers and enterprises access to text generation, reasoning, vision and multimodal, embedding, and speech models through a unified API. Use the console to test models, manage API keys, and monitor usage as you move from a prototype to a production application.

​
Get started

Follow these steps to choose a model, make an API call, and monitor usage. If you have already chosen a model, start with step 2.
1

Choose and test a model

Compare models in the console, then test supported models in the Playground to check whether they fit your use case. Test embedding models through the Embeddings API.
2

Create an API key and call a model

Create an API key and use your chosen model’s ID to make your first API call through REST or an SDK. For failed requests, see Error handling.
3

Check usage and performance

  • View usage: Review per-request token usage, cost, and duration.
  • View metrics: Track overall request trends, throughput, and latency to assess performance and cost.

​
What the platform provides

  • A unified API. Call different models through a common interface, with REST requests and streaming responses. Integrate with mainstream SDKs and AI coding tools while reducing the changes needed to switch models.
  • Inference acceleration and autoscaling. In-house inference acceleration and autoscaling support low latency, high throughput, and stable processing of long-context workloads.
  • Model discovery and testing. Filter the model catalog by provider, category, context window, and price, then test supported models in the Playground.
  • Usage and cost controls. The platform supports pay-as-you-go billing. Set spending caps and token-per-minute (TPM) and request-per-minute (RPM) limits for each API key.
  • Usage monitoring and API key management. View token consumption, spending, request details, and performance metrics on the Home, Usage Detail, and Metrics pages. Manage API keys throughout their lifecycle, including their spending caps and rate limits.

​
Choose a model for your task

Start with the capabilities your task requires, then compare response quality, context window, and price.
A model can support more than one capability. For large documents or codebases that must fit in a single request, also check the context window. See Long context.The catalog changes over time. The model IDs above are examples, not a complete list. Check the model details in Models for available models, supported capabilities, context windows, and prices. Use the model ID shown there in API requests.

​
Common use cases

  • Question answering and assistants. Build enterprise knowledge assistants, operations assistants, and customer service bots that answer questions or help with business tasks.
  • Content creation. Develop ideas, generate marketing copy and long-form content, and write structured documents.
  • Coding and agent workflows. Connect models to coding tools or applications for code generation, agentic coding, refactoring, and multi-step development tasks.
  • Semantic search and RAG. Use embedding models to retrieve relevant content, then provide it as context for a text generation model.
  • Vision and multimodal understanding. Describe images, answer questions about visual content, and analyze charts, screenshots, or documents containing text and images.
  • Speech applications. Transcribe audio or generate spoken responses for voice assistants and accessibility tools.
  • Long-context workflows. Include large documents, code excerpts, or relevant conversation history in one request for question answering, summarization, comparison, or code analysis.

​
Manage usage and costs

Charges depend on actual usage and the selected model’s unit prices. Token-based models are billed by token usage. Speech models are billed by audio duration: input audio for ASR and output audio for TTS. For billing definitions and cost controls, see Token usage and billing.
  • Match the model to the task. Start with a lower-cost model during development, then switch to a more capable model if needed. Compare quality, latency, and cost before making the change.
  • Set limits per application. Create separate API keys for different applications and configure spending caps and TPM/RPM limits to reduce the risk of unexpected usage.
  • Use request data to guide changes. Identify the main sources of usage and cost from request records and aggregate metrics. Adjust the model, input length, or generation budget accordingly.