Why activate these separately

Official free tiers are usually not one lump sum; they are metered per product line. Text chat, vision, speech, and embeddings each have their own quota and rate limits. Many developers only enable chat and leave the rest sitting idle in the console.

This guide only covers switches that actually exist in official consoles, not third-party aggregators.

Steps

Step 1: List every product line available to the account

Sign in to the provider console and open the API keys / usage page. Look at usage broken down by model or endpoint. Common separately enabled categories include:

  • Text chat (chat / completions)
  • Vision (image input)
  • Speech-to-text and TTS
  • Embeddings
  • Some providers also offer rerank, OCR, moderation

If an item shows "not enabled / request access", it is separate from chat quota and needs its own activation.

Step 2: Enable each and record its limits

For each item you enable, note three numbers: free quota (usually refreshed monthly), requests per minute (RPM), and tokens per minute (TPM). Limits vary widely; embedding TPM is often more generous than chat and suits batch indexing.

Step 3: Verify with a minimal request

Do not wire it into production first. Send one minimal request:

  • Chat: send "ping" and confirm a 200 with content
  • Vision: send a small image plus a short prompt
  • Embedding: convert one sentence and confirm the dimension matches the docs
  • Speech: send a few seconds of audio and confirm non-empty transcription

Only integrate after it passes.

Step 4: Assign by use case, do not mix

  • Use embedding quota for knowledge-base indexing, dedup, clustering: one-time batch plus low-frequency increments, saving chat quota.
  • Use vision quota for screenshot understanding and receipt parsing, on demand.
  • Use speech quota for meeting transcription; note it is often billed by minutes, not tokens.

Running these lines separately gives you several non-competing supply lines under one account.

Step 5: Put the refresh cycle on a reminder

Free quotas usually refresh monthly or daily; check the console for the exact rule. Mark the refresh date in a calendar or script so you neither forget to use it at the start of the month nor discover leftovers at the end.

Caveats

  • Free quotas often carry restrictions such as non-production only, region limits, or specific model versions. Read the terms before enabling.
  • Multimodal and embedding billing units differ (images per image, speech per minute, embeddings per token). Do not estimate with chat token counts.
  • Some providers require a payment method to enable certain product lines even with a free quota.
  • After the free quota is exhausted, billing may switch to pay-as-you-go automatically. Set usage caps or budget alerts in the console.
  • Quotas and rate limits change with provider policy. This article avoids hard numbers; rely on the live console.

Use cases

  • Building a RAG knowledge base that needs heavy embeddings but little chat
  • Building screenshot or document understanding tools where vision is the main call
  • Transcribing meetings or podcasts where speech quota is essential
  • Spreading out free supply lines on one account without extra spend