best practices
Recommended solutions for efficient use of Cova Nexus
Model selection strategy
Choosing the most appropriate model based on the scenario can optimize costs while ensuring quality:
| scene | Recommended model | Reason |
|---|---|---|
| Daily conversations/customer service | claude-haiku-4-5 / gemini-3-flash | Low cost, quick response |
| Code generation/review | gpt-5.3-codex-spark / codex-auto-review | Optimized for code scenarios |
| Complex reasoning/analysis | claude-opus-4-6 / gpt-5.5 / grok-4.5 | Strongest reasoning ability |
| Multi-agent/search | grok-4.20-multi-agent-0309 / gemini-pro-agent | Oriented to Agent and multi-agent orchestration |
| Long document processing | gemini-3.1-flash-lite / claude-sonnet-4-6 | Supports very long contexts |
| Image/video generation | gpt-image-2 / gemini-3.1-flash-image / grok-imagine-image / grok-imagine-video | Professional image and video generation capabilities |
cost optimization
ℹ️ Cost optimization suggestions
- Use cheaper models like grok-3-mini (input $0.3/1M) or gemini-3.1-flash-lite (input $0.25/1M) for non-critical requests
- Utilize caching mechanism to reduce token consumption of repeated calculations
- Set max_tokens appropriately to avoid unnecessary output overhead
- Use streaming output to process the first token after obtaining it, reducing the perception of waiting time.
Reliability recommendations
To ensure the stability of the production environment, the following strategies are recommended:
Retry strategy
Retries using exponential backoff for 5xx errors and timeouts, it is recommended to retry up to 3 times with an initial wait of 1 second.
Downgrade plan
Prepare backup models. If gpt-5.5 is unavailable, you can downgrade to gpt-5.4 or claude-opus-4-6.