Billing model

Transparent, per-request pricing. We take 40% of your savings. You keep 60%. STONE saves 0% → you pay 0.

Pricing formula (H.1)

We charge list-price for what actually reaches the provider, plus 40% of the gross savings STONE creates. The customer always nets 60% of the savings.

provider_cost      = actual_input_tokens × P_in + output_tokens × P_out
gross_savings      = max(0, (original_tokens − actual_input_tokens)) × P_in
customer_savings   = gross_savings × 0.60
lexico_profit      = gross_savings × 0.40
customer_charge    = provider_cost + lexico_profit   (credit customer)

// P_in, P_out are list prices per 1M tokens for the selected model.

Turn 1 is always passive

The first request in a new conversation (turnNumber = 1) has no accumulated context to restructure. STONE forwards your payload unchanged and the bomb-proof first-turn guard forces gross_savings = 0, lexico_profit = 0. You pay provider_cost only.

Passive mode

When STONE cannot reduce tokens (e.g. a short single-turn prompt, or a payload that grew after restructuring), gross_savings is clamped to 0 and no Lexi fee applies. The response carries X-Lexi-Mode: passive.

Provider errors

Streaming abort

If your client disconnects mid-stream, billing runs on the tokens actually delivered. The invariant customer_charge ≥ provider_cost always holds. You are never undercharged for what the provider produced, never overcharged for what you didn't receive.

invariant: customer_charge ≥ provider_cost   (always)

Refunds

EU/EEA consumers have 14 days withdrawal right from signup. Manage refund requests at lexico.no/refund. Usage consumed before the refund request is non-refundable per EU Directive 2011/83/EU Art. 16(m).

Rate limits per tier

Limits auto-scale with lifetime spend. No tiers to pick.

Response headers (16)

All 16 canonical X-Lexi headers are exposed on every /v1/chat/completions and /v1/messages response. Full type and unit reference in the headers section of /docs.