AICredits.in Review: 400+ AI Models in INR With UPI Top-Up (What Is Verified)

Payment is the first wall Indian developers hit with AI APIs: the big providers want an international card, and many Indian cards fail. AICredits.in is a gateway built around that problem. You top up a rupee wallet with UPI, net banking or a debit card, and call many providers' models through one OpenAI-compatible endpoint.

This review was rewritten on 2026-10-16. The earlier version read like a hands-on diary, and I cannot back that up, so I replaced it with what I can check: the public model catalog, live requests that need no key, and the homepage text. I had no AICredits API key, so I have not made an authenticated call. Where the claim is the vendor's, I say so. AICredits links on this site are affiliate links.

What is AICredits?

An OpenAI-compatible API gateway. The homepage's own summary: change base_url and api_key, and OpenAI-style clients work. The documented base URL is https://api.aicredits.in/v1.

What I verified without a key:

• GET https://api.aicredits.in/v1/models returns a public catalog. On 2026-10-16 it had 432 models, all marked active: 378 chat, 24 image, 22 embedding, 6 audio and 2 text-to-speech. The full breakdown is in the catalog analysis.
• Calling the chat endpoint with a missing or wrong key returns HTTP 401 with a JSON error ("Invalid API Key" or "Missing or invalid Authorization header").

The homepage counts the models inconsistently ("300+" in some places, "400+" in others). The catalog says 432, so trust the endpoint.

How do you set it up?

Per the homepage: create an account by email, add credits (top-ups go through Razorpay, so UPI, net banking and debit cards work, per the page), create an API key in the dashboard, and replace the base URL. The homepage says credits are valid for one year.

The minimum top-up is unclear. The homepage says "Minimum ₹50 top-up" in two places, and its steps say "Add ₹100+ via UPI". Older posts on this site, including the previous version of this review, said ₹100. Check the dashboard.

What does the code change look like?

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-aicredits-key",
    base_url="https://api.aicredits.in/v1",
)

response = client.chat.completions.create(
    model="openai/gpt-6-luna",   # provider prefix, copied from the catalog
    messages=[{"role": "user", "content": "What's the capital of Rajasthan?"}],
)
print(response.choices[0].message.content)

The code is the standard OpenAI client, and I tested equivalent calls against a local mock, not against AICredits (details). The TypeScript version is the same with baseURL; see the TypeScript and Next.js guide.

Which model IDs should you use?

This is where the earlier version of this review was wrong. Several model names it listed do not exist in the catalog and would return a model-not-found error. Here is what the catalog actually contains on 2026-10-16 for the models that article mentioned:

Do not guess. Use the model ID lookup script, or map your existing names with the migration checklist.

What does it cost?

The homepage says pricing is the "live forex rate + 5% buffer" and, in small print, mentions a 5% platform fee. The public catalog gives per-token prices in USD and INR. I checked all 413 priced models: the INR price equals the USD price times the listed exchange rate (95.10014 that day) times 1.05. Prices per million tokens, from the catalog:

I compared five Anthropic USD prices with Anthropic's own pricing page on the same day and they matched. The earlier version of this article estimated a 10 to 11 percent markup and showed higher rupee figures. The catalog's markup is 5 percent over the exchange rate. Whether your wallet debit adds a further platform fee is the open question, which one real charge would settle; see the cost-per-feature post.

What does the homepage promise, and what should you test?

These are vendor claims. I did not test them.

• Automatic failover: "If a provider returns an error or rate-limits your request, we retry on the next available route automatically." Test it with the approach in retries, timeouts and fallback, and keep a client-side policy anyway.
• Per-key budgets and team allocation: per-key spend caps, credit allocation by email with a rupee limit per person, and low-balance alerts. Try them on a small wallet first.
• Same streaming and tool calling as OpenAI: run the conformance test on your exact model.
• INR billing: the page promises a rupee wallet and no international card fees. I did not find a statement on GST invoices in what I read, so ask support if you need them for accounting.

What are the trade-offs?

• An extra hop and a dependency. If the gateway is down, so is your app, unless you keep a direct fallback. I did not measure latency; the benchmark method shows how to.
• A markup over direct prices, at least 5 percent per the catalog.
• Prepaid, expiring credits. The homepage says credits are valid for one year, so top up what you will use.
• Feature lag. Newer provider features may not be available at once. For example, the catalog lists the responses API only for some OpenAI models, and not for the newest ones (details).
• Your prompts go through another company. I did not read its data terms; read them before sending sensitive data.

Who should use it?

If you are an Indian developer or team without an international card, a rupee wallet and UPI top-up remove the biggest practical obstacle, and a public catalog with INR prices makes budgeting straightforward. If you already have working direct API access, the case is weaker: you would pay a markup for convenience.

Compare options in AICredits vs OpenRouter, or start small: create an account (affiliate link), add a small amount, run the conformance test on one cheap model, and check your first real charge against the catalog price.