Model selection

Choose the model by the demands of the task, by the expected experience and by the acceptable cost. HiMax prioritizes capability, HiNova balances quality and speed, and HiGenesis prioritizes efficiency at high volume.

All three use POST https://api.hinow.ai/v1/chat/completions and the same message format. Start with the recommendation closest to your case and validate the choice with real examples from your application.

HiMax — maximum capability

hinow/himax is the choice for work where errors, omissions or shallow analysis have the greatest impact. It suits problems with several constraints, code that requires architectural context and agents that need to plan before acting.

Use HiMax for:

• Complex reasoning and decisions supported by several pieces of evidence
• Software architecture, review and refactoring
• Careful analysis of contracts, reports and technical documents
• Multi-step agents, with tools and result validation

Trade-off: higher cost per token. Use it where the quality gain can be measured or has direct business value.

HiNova — balance for production

hinow/hinova offers a balance between capability, speed and cost. It is the recommended starting point for most user-facing applications and for teams that are still defining their evaluation criteria.

Use HiNova for:

• Customer service, support and productivity assistants
• Copilots that explain, suggest or transform content
• Text generation, review and summarization
• Interactive APIs where latency shapes the experience

Trade-off: exceptionally complex tasks may benefit from HiMax; very simple, high-volume workloads may be cheaper on HiGenesis.

HiGenesis — efficiency at scale

hinow/higenesis was designed for well-defined tasks executed many times. It is the lowest-cost option in the family and accepts text or image as input, according to the modalities published in the API catalog.

Use HiGenesis for:

• Classification, triage and routing
• Field extraction and normalization
• Summaries, labels and batch transformations
• Image input in automation workflows
• Auxiliary steps of larger pipelines

Trade-off: ambiguous instructions or tasks with many decisions may require a more explicit prompt or a switch to HiNova or HiMax.

Specialized models

For tasks with a format of their own, the family has dedicated models:

• hinow/himegia — creates and edits images from a text description: legible text in the scene, consistent character across images. See Image generation.
• hinow/hivision — dedicated image input: scanned documents, screenshots, charts.
• hinow/hicode — code generation and transformation in automated pipelines. See Code generation.

For vectors, see Embeddings and select a model whose endpoint is /v1/embeddings in your account catalog.

From task to model

Combine models in the same product

A single product can use more than one model. For example: HiMax for the hardest analysis, HiNova for the conversation with the user and HiGenesis to classify or extract data behind the scenes. The model field can be set per route, per feature or per workflow step.

Avoid selecting on price alone or on an isolated demo. Keep an evaluation set with real inputs, objective criteria and edge cases. Compare quality, latency, tokens and total cost before promoting a configuration to production.