One API key for every AI model.

Build coding agents, automations, support bots and document tools on 47 models, including Claude, GPT, Gemini and DeepSeek. OpenAI- and Anthropic-compatible, paid in USD or BDT, with a spending cap on every key.

Get your API key Plans from ৳90 / week

Get $1 free credit when you sign up. No card needed.Use it within 14 days on selected models.

One API keytok_live_••••

47 models from 17 familiesUp to 1.1M tokens of context. Compare them

No international card needed
Setup in one command
Prompts never stored

Example request

DeepSeek V4.1 Flash

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://tokens.bd",
    "ANTHROPIC_AUTH_TOKEN": "tok_live_your_key",
    "ANTHROPIC_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek/deepseek-v4.1-flash",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek/deepseek-v4.1-flash",
    "CLAUDE_CODE_SUBAGENT_MODEL": "deepseek/deepseek-v4.1-flash"
  }
}

order_agent.py

from openai import OpenAI

client = OpenAI(base_url="https://tokens.bd/v1", api_key="tok_live_your_key")
run = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    messages=[{"role": "user", "content": order_event}],
    tools=[get_order, check_stock, send_sms, add_row],
)

support-bot.js

const res = await fetch("https://tokens.bd/v1/chat/completions", {
  method: "POST",
  headers: { Authorization: "Bearer tok_live_your_key" },
  body: JSON.stringify({
    model: "deepseek/deepseek-v4.1-flash",
    messages: [{ role: "system", content: SHOP_FAQ }, ...chat],
  }),
});

extract_invoice.py

client = OpenAI(base_url="https://tokens.bd/v1", api_key="tok_live_your_key")
data = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    response_format={"type": "json_object"},
    messages=[{"role": "user", "content": PROMPT + invoice_text}],
)

› Add retry with backoff to fetchModels()

Read lib/models.ts and lib/http.ts
Edit fetchModels(): 3 retries, 250ms to 2s backoff
Add tests for timeouts and 5xx responses
Run tests: 14 passed

Input tokens: 0
Output tokens: 0
Cost: ৳0.00
In USD: $0.0000

Claude3 models
GPT3 models
Gemini2 models
DeepSeek5 models
Qwen7 models
Kimi4 models
GLM4 models
Grok2 models
MiMo4 models
Muse3 models
MiniMax2 models
Step2 models
Tencent2 models
Inkling1 model
Ling1 model
Longcat1 model
Nemotron1 model

Works with the coding agents and tools you already use

See all 17 setup guides

Trusted & Featured

Connected in three steps

No new SDK and no code changes. Tokens speaks the API formats your tools already use, so connecting is a base URL and a key.

1
Create an account and add credit
Sign up with email or Google, then pick a plan or top up your wallet with bKash, Nagad, Visa Card, Master Card, Rocket, AMEX, DBBL Nexus, Google Pay, Internet Banking and SSL. Usage is metered per token at the prices on the model catalog.
2
Connect your agent
One command signs you in through the browser and configures Claude Code, Codex CLI, OpenCode and Crush, backing up each file first.
terminal
curl -fsSL https://tokens.bd/cli/tokens.mjs -o tokens.mjs node tokens.mjs setup --base-url https://tokens.bd
3
Code with spending under control
Give each key a monthly cap and a list of allowed models. Email alerts warn you at 50, 75 and 90 percent of your allowance.

Plans for daily coding and long agent runs

Choose a plan with a credit allowance, or pay as you go from a prepaid wallet. Prices are set separately in taka and dollars.

Weekly 5X

Starter

৳90/ week

$10 in credits included

Up to ~$20 usage with deals

What's Included~5K requests

Per-model allowancesEach model can use up to this much of your plan's usage every billing period.

$20 on GLM-5.3 Flash
$15 on DeepSeek V4.1 Flash 1017
$15 on DeepSeek V4.1 Flash

+ 2 more models: Meta, Qwen

$15 on Muse Spark 1.2 Contributor
$15 on Qwen 3.8 27B

Up to 5 API keys
All AI agent support
Most coding tools support
CLI available.
5x equivalent of Claude Code/Codex

Subscribe to Weekly 5X

Monthly Max

Pro

৳490/ month

$50 in credits included

Up to ~$60 usage with deals

What's Included~30K requests

Per-model allowancesEach model can use up to this much of your plan's usage every billing period.

$60 on Claude Haiku 5.5
$60 on DeepSeek V4.1 Flash 1017
$60 on DeepSeek V4.1 Flash

+ 4 more models: Z, Qwen, Xiaomi

$60 on GLM-5.3
$60 on Qwen 3.8 27B
$60 on MiMo V2.6 Flash
$60 on Qwen 3.7 Flash

Up to 15 API keys
All AI agent support
Most coding tools support
CLI available.
20x equivalent of Claude Code/Codex

Subscribe to Monthly Max

Pay as you go

No plan

Add credit to a prepaid wallet and pay per token at the prices in the model catalog.

Every model in the catalog
Spend caps and model lists per key
Top up whenever you need to

Create account

Pay with

Internet Banking

Built for developers in Bangladesh who ship with agents

The parts that matter when your editor sends thousands of requests a day: paying locally, keeping spend in check, switching models, and knowing what is stored.

Pay in taka

Top up with bKash, Nagad, Visa Card, Master Card, Rocket, AMEX, DBBL Nexus, Google Pay, Internet Banking and SSL and get a receipt for each payment. No international card, no per-payment paperwork.

Wallet top-up

৳1,000.00

Paid

Paid with: bKash
Receipt: TK-20431
New balance: ৳1,250.00

bKashNagadVisa CardMaster CardRocket

How paying in BDT works

Spending you control

Monthly caps and model allow-lists per key. A runaway agent loop stops at the cap.

claude-code-laptop

tok_live_••••3f9a

This month৳930.00 of ৳1,500.00

Email alerts at 50%, 75% and 90%. Stops at the cap.

claude-sonnet-5.5deepseek-v4.1-flash

API keys and limits

One key, every model

OpenAI- and Anthropic-compatible on the same key. Switch models by changing one id.

{ "model": "deepseek/deepseek-v4.1-flash" , … }

DeepSeek V4.1 Flash৳30.00 / ৳107.50
Claude Sonnet 5.5৳250.00 / ৳1,250.00
GPT-5.6 Luna৳62.50 / ৳375.00
Gemini 3.8 Flash৳225.00 / ৳1,093.75

Input / output per 1M tokens

Choosing a model

Private by default

We keep what billing needs (model, tokens, cost, latency) and never store prompts or responses. No training on your data.

Time	Model	Tokens in / out	Cost	Prompt
14:02:11	claude-sonnet-5.5	24,180 / 1,460	৳8.27	not stored
14:01:37	deepseek-v4.1-flash	11,902 / 640	৳0.29	not stored
13:58:05	glm-5.3	6,410 / 912	৳1.71	not stored

What we log and what we don't

Every coding model, one price list

Each model is called by a stable id, with its context window and input and output prices in taka and dollars.

	Provider				Available on
Claude Opus 5.5anthropic/claude-opus-5-5	Anthropic	1M	$4.00	$20.00	Monthly · Pay as you go
Claude Sonnet 5.5anthropic/claude-sonnet-5.5	Anthropic	1M	$2.00	$10.00	Weekly · Monthly · Pay as you go
Claude Haiku 5.5claude-haiku-5-5	Anthropic	1M	$0.10	$0.50	Free · Weekly · Monthly · Pay as you go
DeepSeek V4.1 Flash 1017deepseek-ai/deepseek-v4.1-flash-1017	DeepSeek	1M	$0.30	$0.80	Free · Weekly · Monthly · Pay as you go
DeepSeek V4 Flash (latest)deepseek/deepseek-v4-flash-latest	DeepSeek	1M	$0.24	$0.86	Weekly · Monthly · Pay as you go
DeepSeek V4 Flash Vision (exp)deepseek/deepseek-v4-flash-vision-exp	DeepSeek	1M	$0.17	$0.66	Weekly · Monthly · Pay as you go
DeepSeek V4 Pro (latest)deepseek/deepseek-v4-pro-latest	DeepSeek	1M	$0.73	$2.18	Weekly · Monthly · Pay as you go
DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flash	DeepSeek	1M	$0.24	$0.86	Weekly · Monthly · Pay as you go
Gemini 3.7 Flashgoogle/gemini-3.7-flash	Other	1M	$1.80	$9.00	Weekly · Monthly · Pay as you go
Gemini 3.8 Flashgoogle/gemini-3.8-flash	Other	1M	$1.80	$8.75	Weekly · Monthly · Pay as you go
Ling 3.1 Flashinclusionai/ling-3.1-flash	InclusionAI	262K	$0.35	$1.00	Weekly · Monthly · Pay as you go
Inkling Smallinkling-small	InclusionAI	1M	$0.55	$1.32	Weekly · Monthly · Pay as you go
LongCat 2.0meituan/longcat-2.0	Other	1M	$0.33	$1.32	Weekly · Monthly · Pay as you go
Muse Spark 1.2 Contributormeta/muse-spark-1.2-contributor	Other	1M	$0.40	$0.80	Weekly · Monthly · Pay as you go

Showing 12 of 47 modelsFull catalog and comparisons

Estimate your monthly bill before you commit

Pick a model and a monthly token volume. Coding agents send far more input than output, and the estimate accounts for that.

Model47 available · input / output per 1M tokens

Monthly token volume10Mtokens / month

1M25.2M50.4M100M

Assumes 70% input and 30% output tokens (7M in, 3M out).

Pay in USD or BDT
Works with OpenAI and Anthropic compatible tools
Spending limits per API key

On Tokens, per month

$88.00

for Claude Opus 5.5

Input / 1M: $4
Output / 1M: $20

Claude Opus 5.5 has a context window of 1M tokens and is billed by the token, with no monthly minimum.

Estimate only. You are billed for the tokens you actually use at the prices on the model pages.

How Tokens compares

Side-by-side comparisons with other gateways and direct APIs, including where the alternative is the better choice. Competitor prices are dated and sourced.

Tokens vsOpenRouterOpenRouter has the larger catalog, free models and fine routing controls. Tokens lets you pay in BDT and sets spend caps and model lists per key.Read the comparison

Tokens vsDeepSeek APIDeepSeek's own API sets the list price for its models and halves it off-peak. Tokens adds BDT billing and other providers on the same key.Read the comparison

Tokens vsOpenCode ZenZen sells tested coding models at cost plus card fees, and OpenCode Go adds $10 and $40 plans. Tokens bills in BDT with one base URL per API format.Read the comparison

We don't store your prompts

Predictable limits and plain data handling for engineers and agent builders. Read exactly what we log.

Usage metadata only
Prompts and completions stream through the gateway and are not archived. Billing needs model, tokens, cost and latency, and that is what we keep.
Keys you can rotate
Rotate a key in one click and choose: the old secret stops at once (for a leak), or keeps working for 1 or 24 hours while you switch your tools over. Admin access needs MFA and every admin action is logged.
Automatic failover
If the provider serving a model fails or stalls, the request moves to the next one before it reaches you.
Status you can check
Live component health and 90-day uptime from real checks, with incidents posted on the status page.

What developers say

“I use tokens for studying and research work. It tremendously helped me and relieved me of the headache of finding different models while using Bangladeshi currency. The website is well-built and informative too. They respond to requests pretty fast. I wish they would slightly reduce the price.”
M.I.RVerified customer
“This is an ultimate automation fuel for my AI agents.”
Rakibul Hasan
Digital Marketer
“With pro plan it's feels like unlimited with Deepseek v4.1 Flash. Totally value for money”
Mohiuddin
CTO

Guides and articles

All articles

New ModelsClaude Fable 5.2: Price, Benchmarks, and AccessAnthropic launched Claude Fable 5.2 on 9 Nov 2026 at $8/$40 per 1M tokens — 20% cheaper than Fable 5.1 and #1 on the Artificial Analysis Index.

GuidesGemini API Pricing 2026: Every Model, Free Tier, CostGemini API pricing for every model, checked 9 November 2026: Argon doubles on 1 Dec, 3.8 Flash free tier, and what a coding session costs in taka.

New ModelsGPT-6 Cyber: Price, Access, and What It Can DoOpenAI launched GPT-6 Cyber on 5 Nov 2026 at $20/$100 per 1M tokens, but access is gated behind Daybreak Red. Price, benchmarks, alternatives.

Questions developers ask before they switch

Something else? The docs cover setup for every agent, and you can reach support from your dashboard.

What is Tokens?

Tokens is an AI model gateway for developers. One API key gives you OpenAI-compatible and Anthropic-compatible endpoints for coding models from several providers, metered per token and paid in USD, BDT.

Can I pay for AI APIs in Bangladeshi taka?

Yes. Prices are shown in USD, BDT and you can pay with bKash, Nagad, Visa Card, Master Card, Rocket, AMEX, DBBL Nexus, Google Pay, Internet Banking, SSL. You do not need a dual-currency international card, and every payment gets a receipt in your dashboard.

Which coding agents work with Tokens?

Claude Code, Codex CLI, OpenCode, Cursor, Cline, Kilo Code, Continue, Zed, GitHub Copilot, OpenClaw, Hermes Agent, Aider, Crush, Goose and any other tool that accepts a custom OpenAI- or Anthropic-compatible base URL. The setup CLI configures Claude Code, Codex CLI, OpenCode and Crush in one command, and the docs have a guide for each tool.

Is the API OpenAI-compatible and Anthropic-compatible?

Both. Use https://tokens.bd/v1 with OpenAI SDKs and tools (chat completions, the Responses API, streaming and tool calling), or https://tokens.bd with Anthropic SDKs and Claude Code. The same API key works on both.

Which models can I use?

Every active model in the catalog at /models, with its context window and price per million tokens in USD and BDT. Your key can also list the models it may call with GET /v1/models.

Do you store my prompts or code?

No. We record usage metadata only (model, token counts, cost, latency and time) for billing. Prompts and responses pass through to the model provider, whose own data policy applies, and we never train on or sell your data.

What happens when my balance or plan allowance runs out?

Requests stop with a clear error instead of running up a bill: 402 when the balance is used up and 429 when a plan usage window is used up, with Retry-After telling you when it resets. Email alerts warn you at 50, 75 and 90 percent first.

Can I limit how much one API key can spend?

Yes. Give each key a monthly spending cap and, optionally, a list of allowed models when you create it. A runaway agent loop then stops at the cap instead of draining your balance.

How is Tokens different from OpenRouter?

OpenRouter has a larger catalog and charges in US dollars by card or crypto. Tokens focuses on coding models and adds what developers in Bangladesh need: payment in taka, per-key spend caps and usage alerts, a one-command agent setup, and support tickets in the dashboard. The full comparison is at /vs/openrouter.

How do I move from another gateway or a direct provider API?

Change the base URL and API key in your tool or SDK, then use model ids from /models (they are written as provider/model). Most tools need nothing else.

How do I get a receipt or ask for a refund?

Receipts for every payment are in Dashboard > Billing and can be printed. Refund requests are reviewed under the refund policy; open a support ticket from your dashboard with the receipt number.

Is there an uptime guarantee?

There is no published SLA yet. The status page at /status shows live component status and 90-day uptime measured from real health checks, and incidents are posted there.

Your first request is five minutes away

Create an account, add credit in USD or BDT, and run one command to connect your coding agents.

Create account See pricing

A short email when new coding models arrive, prices change or we ship something that saves you tokens.

One API key for every AI model.

Works with the coding agents and tools you already use

Connected in three steps

Create an account and add credit

Connect your agent

Code with spending under control

Plans for daily coding and long agent runs

Weekly 5X

Monthly Max

Pay as you go

Built for developers in Bangladesh who ship with agents

Pay in taka

Spending you control

One key, every model

Private by default

Every coding model, one price list

Estimate your monthly bill before you commit

How Tokens compares

We don't store your prompts

Usage metadata only

Keys you can rotate

Automatic failover

Status you can check

What developers say

Guides and articles

Questions developers ask before they switch

What is Tokens?

Can I pay for AI APIs in Bangladeshi taka?

Which coding agents work with Tokens?

Is the API OpenAI-compatible and Anthropic-compatible?

Which models can I use?

Do you store my prompts or code?

What happens when my balance or plan allowance runs out?

Can I limit how much one API key can spend?

How is Tokens different from OpenRouter?

How do I move from another gateway or a direct provider API?

How do I get a receipt or ask for a refund?

Is there an uptime guarantee?

Your first request is five minutes away

Stay ahead of every model release

What developers say