How token price works

Live DevShard escrows are 10 ngonka per token per attempt. The old governance endpoint still says 1. OpenBroker adds no markup.

On this page

A usage line is not 1 ngonka per token, even though GET /v1/governance/pricing still returns price_per_token: 1 for DeepSeek and MiniMax. OpenBroker does not add a commission. The difference is the live DevShard protocol.

Three numbers people mix up

SourceRateWhat it is
/v1/governance/pricing1 ngonka / tokenThe pre-DevShard DAPI price (0.001 GNK per 1M tokens). Still published. Not what escrows pay.
devshard_escrow_params.token_price10 ngonka / tokenStamped onto every escrow when it is created. One successful attempt costs (prompt + completion) × 10 ngonka.
Typical billed lineabout 15 ngonka / tokenThe gateway races or retries the same request. Each attempt burns tokens × 10. One to about 2.5 attempts is common, so the effective rate is usually 10–25 ngonka per token.

Per-attempt math

The gateway reserves the worst case, then settles the tokens that actually ran. Surplus returns to the escrow lock. It is not a separate refund line on your broker ledger.

text
reserved = (input_length + max_tokens) × token_price
actual   = (prompt_tokens + completion_tokens) × token_price
# token_price = 10 ngonka on mainnet DevShard escrows

Why the usage report shows about 15 ngonka per token

When the gateway does not return a per-request settled cost, OpenBroker estimates tokens × 10 × 1.5 and labels the row estimated_token_price_with_retries. The 1.5× factor is the observed average redundancy burn, not a markup. If the gateway later reports all_attempts_actual_cost, that row is upgraded to the settled amount.

Example: 9,898 tokens billed as 148,470 ngonka is 9898 × 10 × 1.5, not 9,898 × 1 from governance pricing.

Broker markup_bps is 0. Confirm the live rate with chain inference params (token_price) versus the stale governance pricing endpoint.

Protocol references: escrow create stamps token_price, reserved and actual cost, gateway redundancy.