Table of contents

Pricing & Usage

Podstack Inference is pay-as-you-go, billed per token and deducted directly from your Podstack wallet — there’s no subscription and no separate invoicing step. You pay for the input (prompt) and output (completion) tokens you actually use, at a per-model rate.

How billing works

  1. Pre-request check. Before a request runs, the gateway verifies your wallet has a positive balance. If it’s zero or negative, the request returns 402 insufficient_funds and nothing is charged.
  2. Token counting. The serving engine reports prompt_tokens and completion_tokens in the response usage object. Those counts drive the charge.
  3. Cost = tokens x per-model rate. Each model has an input rate and an output rate, priced per 1M tokens (in INR by default).
  4. Wallet debit. The computed cost is deducted from your wallet, attributed to the request and model.

Cache hits are free. If you enable the per-key response cache and a request is served from cache, it is recorded with zero cost and doesn’t count against your rate or token limits.

Check pricing

Pricing is available from a public endpoint (no auth required):

curl https://cloud.podstack.ai/infer/v1/pricing

{
  "models": [
    { "model": "...", "input_per_1m": "...", "output_per_1m": "...", "currency": "INR" }
  ]
}

You can also see each model’s input/output rate on its card in the Inference > Catalog view. See Models.

The wallet

Your wallet balance funds all inference usage. In the portal:

  • The Usage view shows your current balance (in ₹) with a Top up button and a low-balance banner.
  • Top-ups are processed through the portal (card/UPI via Razorpay, or PayPal), with preset amounts and a minimum top-up.

Keep a buffer above zero — a wallet at or below zero blocks new requests with 402 insufficient_funds.

Usage analytics

Track spend and performance under Inference > Usage, or over the API.

Summary

GET /v1/usage/summary returns aggregate metrics:

  • total_requests, cached_requests
  • total_tokens, prompt_tokens, completion_tokens
  • total_cost
  • avg_latency_ms, avg_ttft_ms (time to first token)

Per-request history

GET /v1/usage/requests (supports limit / offset) returns individual requests with time, model, input/output tokens, cost, latency, and status (including whether a response was a cache hit).

curl "https://cloud.podstack.ai/infer/v1/usage/requests?limit=20" \
  -H "Authorization: Bearer $PODSTACK_API_KEY"

Use these to attribute spend across teams and keys, spot runaway clients, and pick the right model for your latency and cost budget.

Controlling spend

  • Set a monthly token limit per key so a single client can’t overspend. See Authentication.
  • Enable the response cache on keys with repetitive prompts — cache hits are free.
  • Prefer smaller models where quality allows; the per-token rate scales with model size.
  • Cap max_tokens to bound the output (and cost) of each request.

Next steps