RelevancyLens runs on OpenRouter, so you can use a different model for each task. Match cheaper, faster models to simple jobs and stronger models to reasoning-heavy ones to balance quality, speed, and cost.
Browse all OpenRouter modelsKnowing the call pattern helps you pick where a stronger model is worth it
Three common ways to balance quality against cost
High-volume audits where throughput matters more than nuance.
A capable default model, upgraded only where reasoning pays off.
Client deliverables where quality of insight is the priority.
RelevancyLens ships with an advanced open-weight model so you get
near-frontier quality for roughly a tenth of GPT-4o's price. Override any task in
Admin → Models.
| Role | Default model | Why |
|---|---|---|
| All chat tasks | deepseek/deepseek-chat DeepSeek V3 · open weight |
GPT-4-class reasoning & JSON output; ~10× cheaper than GPT-4o |
| Embeddings | openai/text-embedding-3-small |
Already ~$0.02 / 1M tokens; scoring is calibrated to it |
Strong open-weight alternatives (also cheap):
Each task can use its own model — set your own under Model Preferences
| Task | Type | What it does | Model tip |
|---|---|---|---|
| Primary Topic Discovery | chat | Finds the page's core topic | Fast model is fine |
| Intent Analysis | chat | Generates likely user questions | Mid-tier model |
| Gap Analysis | chat | Finds missing but relevant topics | Stronger model helps |
| Entity Analysis | chat | Existing + high-value missing entities | Stronger model helps |
| Chunk Rewriting | chat | Rewrites content for relevance | Quality model for client work |
| Metadata Improvement | chat | Improves title / description / H1 | Mid-tier model |
| Embeddings / Relevancy | embedding | Scores chunk relevancy | Cost-efficient embeddings |
An administrator chooses the model for each task from the admin dashboard. Changes apply to everyone using the app — no redeploy needed.
Add your OpenRouter key, choose your models, and start analyzing.