TL;DR: You don't need a US card to call Claude — but only two routes truly need no card at all: a BDT gateway like Tokens (bKash/Nagad top-ups, Sonnet 5.5 at $2.20/$11.00 per 1M, ~10% over Anthropic list) or crypto credits on OpenRouter / AI/ML API. The cloud routes (Bedrock, Vertex AI) don't remove the card requirement — they move it to whoever owns the cloud account. On price, Bedrock, Vertex, and OpenRouter all bill Sonnet 5.5 at Anthropic's $2/$10 list; AI/ML API marks up ~30% ($2.60/$13). All prices checked 18 October 2026.
Last week we covered why Bangladeshi cards get declined on Anthropic's console and the payment mechanics behind it. This post answers the next question: which access route should you actually use? Not how billing works — which door to walk through, what it costs versus Anthropic's list price, how much setup each one needs, and who each route is really for.
The five routes at a glance#
Anthropic list prices, checked 18 October 2026 (USD per 1M tokens):
| Model | Input | Output |
|---|---|---|
| Haiku 5.5 (prompts ≤ 100K) | $0.10 | $0.50 |
| Haiku 5.5 (prompts > 100K) | $0.50 | $2.50 |
| Sonnet 5.5 | $2.00 | $10.00 |
| Opus 5.5 | $4.00 | $20.00 |
| Fable 5.1 | $10.00 | $50.00 |
List prices per Anthropic's pricing page; Haiku 5.5's lower tier applies to prompts up to 100,000 tokens.
| Route | No US card? | Sonnet 5.5 price vs list | Claude models available | Setup effort |
|---|---|---|---|---|
| Tokens (BDT gateway) | Yes — bKash/Nagad | $2.20/$11.00 (+10%) | Sonnet 5.5 (catalog checked 18 Oct 2026) | Minutes |
| OpenRouter | Yes — crypto | $2/$10 (list) + 5.5% card fee | Full current lineup incl. Opus 5.5, Haiku 5.5 | Minutes |
| AWS Bedrock | Via your org's AWS account | $2/$10 (list) | Opus 5.5, Sonnet 5.5, Haiku 5.5 | Hours |
| Google Vertex AI | Via your org's GCP account | $2/$10 (list, global) | Opus 5.5, Sonnet 5.5 (+ Haiku rolling out) | Hours |
| AI/ML API | Yes — crypto | $2.60/$13 (+30%) | Opus 5.5, Sonnet 5.5, Fable 5.1 | Minutes |
Route 1: Tokens — pay in taka, no card at all#
Tokens is a BDT-first AI gateway: you top up with bKash (Send Money / Payment) or Nagad (Send Money / Merchant), get a receipt per payment, and call the models through an OpenAI-compatible endpoint. No international card touches the flow at any point — this is the only route here where the card question never comes up.
Price vs list: Claude Sonnet 5.5 is listed at $2.20/$11.00 per 1M tokens on the live catalog (checked 18 October 2026) — about 10% over Anthropic's $2/$10 list. Prices are quoted separately in taka and dollars, so the ৳ figure is fixed and never drifts with the exchange rate. At ৳123.54 per USD (18 October 2026), that's roughly ৳272/৳1,359 per 1M tokens.
Model coverage (honest note): as of 18 October 2026 the catalog carries Claude Sonnet 5.5 only — no Opus 5.5 or Haiku 5.5 yet. If you specifically need Opus or Haiku, use routes 2–5 below and check the live model list before you commit.
Setup: sign up, top up in BDT, create an API key with a monthly spend cap, point your OpenAI SDK at the new base URL. Minutes, and the key-level spend caps mean a runaway agent can't burn more than you pre-approved.
Best for: solo developers in Bangladesh (or anywhere without a US card) who want Claude running today with zero card paperwork.
Route 2: OpenRouter — one key, crypto accepted#
OpenRouter is the best-known aggregator: one API key, hundreds of models, Claude included. It passes through provider per-token prices with no per-token markup — its cut is a 5.5% fee on credit purchases made by card, published on its pricing page.
No-card path: OpenRouter accepts crypto for prepaid credits alongside cards. Buy credits with crypto and no card is ever involved. (Card buyers pay the 5.5% fee; crypto avoids both the card and the fee.)
Price vs list: Sonnet 5.5 at Anthropic's list $2/$10; Opus 5.5 and Haiku 5.5 also at list. Opus 5.5 currently leads OpenRouter's spending share, so availability is solid.
Setup: minutes — account, buy credits, generate a key, OpenAI-compatible endpoint. See the trade-offs against a BDT gateway in Tokens vs OpenRouter.
Best for: developers who want Claude plus every other model on one key and can pay with crypto — or who already have a card and just want model choice.
Route 3: AWS Bedrock — Claude on your AWS bill#
Bedrock serves the current Claude lineup — Opus 5.5, Sonnet 5.5, Haiku 5.5 — at Anthropic's list prices, billed through your AWS account. AWS expanded Claude availability across Asia-Pacific in September 2026 (India, Seoul, Singapore in-region inference), on top of the existing US/EU/APAC regions.
No-card path (the honest version): Bedrock doesn't remove the card requirement — it moves it to whoever owns the AWS account. If your employer or startup already has AWS billing set up, you as a developer never touch a card; you get IAM credentials and call the Converse API or the Anthropic SDK pointed at Bedrock. If you personally have to open the AWS account, you're back to the card wall.
Price vs list: list prices on standard on-demand; note Bedrock regional endpoints can carry premiums — and prompt caching on Bedrock needs a 4,096-token minimum, as covered in the three-provider caching comparison.
Setup: hours — AWS account, IAM role, request model access in the Bedrock console, then code against the Converse API or Bedrock-flavored SDK auth. Not a drop-in base-URL swap.
Best for: teams already living on AWS, especially with data-residency needs (in-region inference keeps prompts inside the region you call).
Route 4: Google Cloud Vertex AI — Claude on your GCP bill#
Same shape as Bedrock, Google flavor: Claude on Vertex AI (Model Garden) bills at Anthropic's list rates on the global endpoint — Opus 5.5 at $4/$20, Sonnet 5.5 at $2/$10 — with Google listing a second regional rate card roughly 10% higher.
No-card path: identical honesty as Bedrock — the card belongs to the GCP billing account owner, not necessarily to you. Works great through an employer; doesn't solve the solo-dev card problem.
Setup: hours — GCP project, billing account, service account with Application Default Credentials, then the Anthropic SDK in Vertex mode.
Best for: teams already on Google Cloud who want Claude next to Gemini under one bill and one IAM system.
Route 5: AI/ML API — the aggregator with a listed markup#
AI/ML API (aimlapi.com) is a straightforward aggregator: OpenAI-compatible chat-completions endpoint, model IDs like anthropic/claude-sonnet-5.5, and the current Claude 5.5 lineup live — including Opus 5.5 and Fable 5.1, which the BDT gateway doesn't carry.
No-card path: card or crypto for credits — crypto needs no card.
Price vs list (verified on their model pages, checked 18 October 2026): this route charges a visible ~30% markup over Anthropic list:
| Model | Anthropic list | AI/ML API | Markup |
|---|---|---|---|
| Sonnet 5.5 | $2/$10 | $2.60/$13 | +30% |
| Opus 5.5 | $4/$20 | $5.20/$26 | +30% |
| Fable 5.1 | $10/$50 | $13/$65 | +30% |
No hidden fees beyond that — the markup is the business model. Whether it's worth it depends on your volume (below).
Setup: minutes — account, buy credits, key, OpenAI-compatible base URL.
Best for: developers who want Opus 5.5 or Fable 5.1 through a simple OpenAI-style endpoint, without opening a cloud account, and will pay ~30% for the convenience.
Worked example: 10M input + 2M output tokens on Sonnet 5.5#
Same workload, every route. (Output excluded from caching math — none of these routes discount output.)
| Route | Math | USD | BDT (at ৳123.54) |
|---|---|---|---|
| Anthropic list (reference) | 10 × $2.00 + 2 × $10.00 | $40.00 | ৳4,942 |
| Bedrock (standard) | same as list | $40.00 | ৳4,942 |
| Vertex AI (global) | same as list | $40.00 | ৳4,942 |
| OpenRouter | $40.00 + 5.5% card fee | $42.20 | ৳5,213 |
| Tokens | 10 × $2.20 + 2 × $11.00 | $44.00 | ৳5,436 |
| AI/ML API | 10 × $2.60 + 2 × $13.00 | $52.00 | ৳6,424 |
Arithmetic: $40 × 123.54 = ৳4,941.60; $42.20 × 123.54 = ৳5,213.39; $44 × 123.54 = ৳5,435.76; $52 × 123.54 = ৳6,424.08. Rate checked 18 October 2026.
The spread is real but not dramatic at Sonnet scale: the cheapest no-card routes (Tokens, OpenRouter-with-crypto) land within ~10% of list. The markup only bites hard at Opus/Fable scale or very high volume — which is exactly when the cloud routes (if your org has the account) earn their keep.
Recommendation matrix#
| Your situation | Use this route | Why |
|---|---|---|
| Solo dev in Bangladesh, no card, want Claude today | Tokens | bKash/Nagad, no card anywhere, spend caps per key |
| Want Claude + 400 other models, can pay crypto | OpenRouter | List prices, one key, no card via crypto |
| Team already on AWS (org owns billing) | Bedrock | List prices, IAM controls, in-region inference |
| Team already on GCP (org owns billing) | Vertex AI | List prices on global endpoint, one bill with Gemini |
| Need Opus 5.5/Fable 5.1, no cloud account, fine with markup | AI/ML API | Minutes to set up, ~30% over list |
| Need Haiku 5.5 specifically, cheapest | OpenRouter / Bedrock | Tokens doesn't list Haiku yet (checked 18 Oct 2026) |
One more lever that stacks with every route: prompt caching cuts repeated-input cost ~90% on all of them — the three-provider caching comparison has the break-even math. And whichever route you pick, cap it: the spend-caps guide shows how per-key monthly limits stop one buggy agent from burning the budget.
Claude access-routes FAQ#
Can I really use the Claude API with no credit card at all? Yes. Tokens takes bKash/Nagad with no card involved, and OpenRouter and AI/ML API both accept crypto for prepaid credits. The cloud routes (Bedrock, Vertex AI) need a card on the cloud billing account — fine if your employer owns it, not a solo solution.
Which route is cheapest for Sonnet 5.5? Bedrock, Vertex AI (global endpoint), and OpenRouter all bill at Anthropic's $2/$10 list. Tokens is ~10% over list but fixed in taka; AI/ML API is ~30% over list. For the 10M/2M workload above: $40.00 (list routes), $42.20 (OpenRouter with card fee), $44.00 (Tokens), $52.00 (AI/ML API).
Do I need to rewrite my code for Bedrock or Vertex AI?
Mostly yes — Bedrock uses the Converse API or AWS-authenticated SDK calls, Vertex uses the Anthropic SDK in Vertex mode. The gateways (Tokens, OpenRouter, AI/ML API) are OpenAI-compatible: one base_url change and your existing code runs.
Why doesn't Tokens list Opus 5.5 or Haiku 5.5? The live catalog on 18 October 2026 carries Claude Sonnet 5.5 only. Catalogs change — check the model list for what's live before you plan around a specific model.
Is my prompt data handled differently across routes? Yes. On Bedrock and Vertex AI, prompts stay inside your own cloud account and region. On aggregators and gateways, requests route through the provider's infrastructure — read each one's data policy. Tokens states private-by-default (prompts never stored); see the docs.
Route prices and availability checked 18 October 2026. Catalogs, fees, and cloud availability change; each provider's pricing page is the authority, and the model catalog for what's live on Tokens.
Sources: Anthropic pricing · tokens.bd models · OpenRouter pricing · AI/ML API Claude Opus 5.5 · AWS Bedrock Claude region expansion · Vertex AI Claude pricing · Claude API pricing Oct 2026 · USD/BDT rate
