📚 Supported AI Models & Cost Estimation

cella's Inference Stats panel (M) tracks token usage and estimates cost for recognized AI models.

Supported Model Families

cella recognizes models from the following providers:

OpenAI / GitHub Copilot — GPT-4o, GPT-4.1, GPT-5-mini, and their variants
Anthropic — Claude Opus, Sonnet, Haiku families
Google — Gemini Pro, Flash families

How Cost Estimation Works

cella maintains a knownPricing map in the source code with per-model input/output token pricing (USD per 1M tokens). When an API response is intercepted, cella:

Extracts the model name from the SSE response stream
Looks up pricing using fuzzy matching (see below)
Multiplies token counts by the per-token rate
Displays cumulative cost in the Inference Stats panel

⚠️ Pricing changes frequently. The values compiled into cella are approximate and may become outdated. Always check the provider's official pricing page for current rates.

Fuzzy Matching

cella uses case-insensitive substring matching. If the intercepted model string contains a known key, the corresponding pricing is applied.

For example, github-copilot/gpt-4o matches the gpt-4o pricing entry.

Unknown Models

If a model isn't recognized, cella still tracks token counts and request rates — it just reports $0.00 for cost. You can identify these in the Inference Stats panel and add pricing to the source code.

Updating Pricing

To update or add model pricing, edit the knownPricing map in internal/proxy/inference_stats.go:

var knownPricing = map[string]ModelPricing{
    "gpt-4o":       {InputPer1M: 2.50, OutputPer1M: 10.00},
    "claude-sonnet-4-5": {InputPer1M: 3.00, OutputPer1M: 15.00},
    // Add new models here...
}

Then rebuild: go build -o cella ./cmd/main.go

← Home