Intelligence Slider
Pick a number between 0 and 100. That's your cost vs. quality tradeoff.
The idea
Most AI platforms make you choose a model upfront. But why should a simple "what time is it?" go to the same model as "design a microservices architecture"?
The Intelligence Slider routes requests based on complexity. Set it low, and simple tasks go to fast/cheap models. Set it high, and you get the reasoning engines with extended thinking.
Tiers
Economy (0-40%)
Fast models for high-volume, simple tasks.
0-20%
Claude Haiku, GPT-4o-mini
21-40%
Claude Haiku, GPT-4o-mini (more context)
Good for:
Q&A
Code formatting
Text processing
Anything you'd run thousands of times
Balanced (41-60%)
The workhorse tier. Good enough for most things.
41-50%
Claude Sonnet, GPT-4o
51-60%
Claude Sonnet, GPT-4o (enhanced reasoning)
Good for:
Code generation
Document analysis
Research
Most day-to-day work
Premium (61-100%)
The heavy hitters. Extended thinking enabled.
61-80%
Claude Opus, GPT-4o
81-100%
Claude Opus, o1 (full extended thinking)
Good for:
Architecture design
Debugging complex issues
Critical business logic
Anything where you'd rather pay more than get it wrong
Extended thinking
Above 60%, models that support it get extended thinking enabled. The model can:
Break down problems step by step
Consider multiple approaches
Catch its own mistakes
Go deeper on analysis
It's slower and costs more. But for the hard stuff, it's worth it.
Setting the slider
In the UI
There's a slider in the chat settings. Drag it.
Via SDK
Via API
Per user
Admins can set defaults per user:
What it costs
Rough numbers:
0-40%
$1
100 simple requests
41-60%
$3-5
50 moderate requests
61-100%
$10-50
10-20 complex requests
Actual costs vary by provider and token usage.
Tips
Start low - See if 30% handles your use case before cranking it up
Match complexity - A formatting task doesn't need Opus
Watch the metrics - The admin portal shows cost breakdowns
Set sensible defaults - Different user tiers can have different defaults
Last updated