For Farsi-speaking developers

Your coding tools work here —
and paying for it is easy.

Real, payable access to frontier models — with tested configuration, prices you can check against the provider's own, and a named Farsi-speaking human when it breaks.

Get startedSee the prices

A free account starts with ۲۰۰٬۰۰۰ تومان of credit — no bank card needed.

# tested configuration
$ export ANTHROPIC_BASE_URL=https://gateway.pilot.banafsh.com
$ export ANTHROPIC_AUTH_TOKEN=sk-••••••••••••
$ claude
✓ connected · gemini-2.5-pro
✓ cache on — cached tokens bill at one tenth
99.8% uptimeillustrativeNothing measures this yet — a public status page follows the pilot.
CheckableEvery model's price sits next to the provider's official one — compare it yourself
Cache works for you۲۱% of platform tokens come from cache right now — the saving is on your invoice
A real humanillustrativeHas a name and a Telegram — response time committed after the pilot

Why Banafsh, not a borrowed account?

Access is only half of it; the other half is that it still works tomorrow and you know what you spent.

Payment

You can genuinely pay

Top up in Toman, by local card or a voucher code. No anonymous middleman. Official invoices for companies.

Setup

Tested configuration

For Claude Code, Cursor and the SDKs — generated from the live service config, so it never goes stale.

Transparency

Every session, with its cost

See what each run cost, set spend caps, and hear about a low balance before it runs out — not after.

Models — with their real prices, side by side

۱۱۵ models available

Every enabled model with its real price; none of them locked behind a pricier plan. In auto mode Banafsh picks the model for you and shows which one answered.

Alibaba۲۰ models
Google۱۴ models
Anthropic۸ models
Meta۶ models
Moonshot۶ models
OpenAI۵ models
Z.ai۵ models
ByteDance۴ models
NVIDIA۴ models
DeepSeek۳ models
Mistral۳ models
Amazon۲ models
MiniMax۲ models
InclusionAI۱ models
Microsoft۱ models
StepFun۱ models
Xiaomi۱ models

🤝 How we choose: The suggestion below comes from each model's capability tags in our catalog (code, long context, writing…), picking the strongest or the cheapest among them. It is not an independent benchmark — when we have an agentic test, we will publish its results here.

Which model for your work?

Pick your task — our recommendation, a cheaper alternative, and the reasoning:

Recommended

gemini-2.5-pro

Best tool-call accuracy over long sessions; with cache, the real cost is far below the list price.

Cheapest

gpt-oss-120b

Close to Sonnet on agentic work for a fraction of the input price. Great for getting started and repetitive work.

Cheaper alternative

glm-4-7-flash

Close to Sonnet on agentic work for a fraction of the input price. Great for getting started and repetitive work.

Agentic sessions are the heaviest workload — the pricing page shows real per-model costs. With a subscription plan, covered models draw from your weekly cap, not your wallet.illustrative

Get started and grab a key
For teams

A company investing in AI?

A seat per developer, budgets per project, monthly official invoices and human Farsi support. Built for teams of 5 to 50.

Talk to us
Chat

Just want to try it first?

Banafsh chat is free to start — a demo and a sanity check. When you're ready, the same account becomes your API key.

Open chat