Beginner question: What's the actual difference between web chat and API behavior?

I've been evaluating DeepSeek Chat for potential integration into a monitoring service I'm building. While testing both interfaces, I noticed behavioral differences that go beyond just the UI. For backend integration planning, understanding the actual divergence between web chat and API responses is crucial.

From my testing:

**Web Chat Behavior:**
- Often includes more verbose explanations by default
- Sometimes adds formatting (like bullet points) even when not explicitly requested
- Appears to have a slightly different temperature setting or system prompt
- May include "safety" disclaimers that the API omits

**API Behavior (using typical config):**
```json
{
"model": "deepseek-chat",
"messages": [{"role": "user", "content": "Explain JOIN types in SQL"}],
"temperature": 0.7,
"max_tokens": 1000
}
```
- Responses tend to be more concise with identical prompts
- Follows the exact parameters provided without "helpful" additions
- More predictable token usage (important for cost estimation)
- No UI-injected formatting unless explicitly requested

Has anyone else done systematic comparisons? I'm particularly interested in:
- Whether there are actual model differences or just parameter defaults
- Consistency in complex technical responses (query optimization explanations, algorithm breakdowns)
- Token efficiency comparisons for similar quality outputs

For my use case—generating database optimization suggestions via API—I need predictable, concise outputs without manual post-processing. The web interface is great for exploration, but the API's behavior seems better suited for programmatic use.

-- latency