pricing

Cost Calculation

ActiveHarness automatically calculates the monetary cost of every LLM call and attaches it to the Result object.

Cost is computed from token usage and per-model pricing data bundled with the gem (sourced from models.dev).

---

Reading Cost from a Result

agent = SupportAgent.call(input: "What is your return policy?")
result = agent.result

if result.usage.cost
  puts result.usage.cost.input   # => 0.00003
  puts result.usage.cost.output  # => 0.00024
  puts result.usage.cost.total   # => 0.00027
else
  puts "Cost unavailable for this model"
end

All values are in USD, rounded to 8 decimal places.

result.usage.cost is nil when:

• The model is not found in the pricing registry.
• The provider did not return usage data (some free-tier models, streaming without usage header).
• Pricing data is missing an input or output price for the model.

---

Pricing Registry

Pricing data is provided by the separate active_harness_pricing gem (a runtime dependency of ActiveHarness, require "active_harness_pricing" — not code that lives in this repo). It fetches model data from models.dev and caches it to {project_root}/tmp/active_harness/models_dev_pricing.json. The cache (and the in-memory copy) is considered fresh for 72 hours; once stale, the next lookup triggers a background refetch automatically.

Force a refresh manually:

ActiveHarness::Pricing.preload!   # re-fetches from models.dev; network failures are silently ignored

There is no public ActiveHarness::Pricing.update — that method only exists on the internal ActiveHarness::Pricing::ModelsDev source (ActiveHarness::Pricing::ModelsDev.update), which raises on failure instead of swallowing it.

Look up a model:

m = ActiveHarness::Pricing.find("mistralai/mistral-nemo")
puts m.input_per_million    # => 0.15   (USD per 1M input tokens)
puts m.output_per_million   # => 0.15   (USD per 1M output tokens)