Skip to content
Last updated

Freddy uses synapses and neurons as its units for measuring AI model computation. Understanding these helps you predict costs and optimize usage.

What Are Neurons?

Neurons measure input processing — the computational work required to read and understand everything you send to the model:

  • Input neurons — Your text messages and prompts
  • Context neurons — Thread history included in the request
  • System neurons — Instructions and system prompts from your assistant

Every token in your request context consumes neurons.

What Are Synapses?

Synapses measure output generation — the computational work required to produce the model's response:

  • Output synapses — The visible text, structured data, or tool calls in the response
  • Reasoning synapses — Internal thinking steps for reasoning models (counted even when not shown)
  • Tool execution synapses — Processing during function calls

Every token the model generates consumes synapses.

Why Two Separate Units?

Input and output computation are fundamentally different workloads. Separating them gives you:

  • Transparency — See exactly where your costs come from
  • Control — Limit output independently from input with max_output_synapses
  • Fairness — Pay for the actual computation performed, not a blended average

Token Conversion

Tokens (used internally by language models) and synapses/neurons have a direct relationship:

  • 1 synapse ≈ 10 output tokens
  • 1 neuron ≈ 10 input tokens

These are rounded estimates. The exact conversion may vary by model.

Pricing

MetricRate
Neurons8.00 CHF per 1M neurons
SynapsesVaries by model

See Pricing for full model-by-model rates.

Response Usage Fields

Every response includes a usage object with token counts:

{
 "usage": {
 "input_tokens": 245,
 "output_tokens": 87,
 "total_tokens": 332
 }
}

Your usage dashboard in Freddy shows the corresponding synapse and neuron counts.

Controlling Output Cost

Use max_output_synapses to cap response length and cost:

{
 "organization_id": "org_your_org_id",
 "model": "gpt-4o",
 "max_output_synapses": 512,
 "inputs": [{"role": "user", "content": "Explain quantum computing briefly."}]
}

Common values:

ValueUse Case
128One-sentence answers
512Short paragraphs
2048Standard responses (default)
4096Detailed explanations
8192Long documents or code

If the model reaches the limit mid-response, the output is truncated.

Monitoring Usage

Track consumption via:

  • Freddy → Organization → Usage dashboard
  • Analytics API — Programmatic access, see the API Reference