Pay only for what you use.

No subscription. The open frontier — Kimi, GLM, MiniMax — sits well under market on every bucket: input, cache, and output. Claude (Anthropic) is pass-through on your own key, never marked up.

Open models, per token

vs OpenRouter · 2026-05-31

Kimi K2.6

71–86% cheaper, end to end

Input$0.20 $0.68−71%Cached$0.020 $0.14−86%Output$0.80 $3.42−77%

GLM 5.1

84–92% cheaper, end to end

Input$0.15 $0.98−85%Cached$0.015 $0.18−92%Output$0.50 $3.08−84%

MiniMax M2.7

25–31% cheaper, end to end

Input$0.18 $0.26−31%Cached$0.036 $0.052*−31%Output$0.90 $1.20−25%

All values in $/M tokens. Strikethrough is the cheapest non-InferRoute access (OpenRouter). Cached is the discounted rate for tokens served from a repeated prefix. * MiniMax M2.7 has no published cache rate, so OpenRouter's is imputed at its typical 20% input ratio.

Works with the Claude you already pay for. Sonnet & Opus pass through your own Anthropic subscription — never marked up, never touched. InferRoute just adds the cheaper open models on top.

Two ways to pay

Most people

One account, no subscription

Top up a balance and pay only for what you route — no monthly bill, stop anytime.

No usage-cap anxiety — never ration prompts or get throttled
Prepaid credits that never expire
Top up by card or USDC — 0% fees in USDC

Get started

x402 enabled

Pay-as-you-go in USDC

Approve once via the open x402 standard. Usage is metered per request and topped up in $1 USDC increments — batched on purpose, so you pay 0 network fees.

$1 gasless top-ups, drawn down per request — 0% fees
Non-custodial: you approve, funds stay yours until used
Built for agents and machine-to-machine spend

Learn about x402

Pay only for what you use. No subscription, no commitment.