5 Affordable Claude API Options in 2026

OpenRouter, AWS Bedrock and Google Vertex AI generally charge around the list price, but bring other strengths: many model families in one account, or billing and compliance inside a cloud you already use. If your prompts repeat a lot, prompt caching lowers the bill on any of them.

1. APIVAI: a prepaid gateway for Claude and GPT models with both Anthropic and OpenAI API formats.
2. OpenRouter: one account for hundreds of models from many providers.
3. AWS Bedrock: Claude inside your AWS account, billed on your AWS invoice.
4. Google Cloud Vertex AI: Claude inside your Google Cloud project.
5. Anthropic API with prompt caching: the official API, made cheaper by caching repeated context.

Official and APIVAI prices per 1M tokens, as of 2026-10-10:

Official prices come from Anthropic pricing; every APIVAI model is on the live pricing page. For OpenRouter, Bedrock and Vertex AI, check OpenRouter pricing, Bedrock pricing and Vertex AI pricing for current prices.

A worked example: a day of Claude Code with Claude Sonnet 4.6 at 20 sessions of 50K input + 20K output tokens costs $9.00 at official prices and $3.45 on APIVAI. The same day on Claude Opus 5.5 costs $12.00 official versus $4.60 on APIVAI.

How pricing works: a prepaid USD balance. Each request is charged for its input, output and cache tokens at the rates on /pricing. There is no subscription and no monthly minimum.

Setup:

1. Sign up with your email and a verification code.
2. Top up from $10 (or ¥20 via Alipay / WeChat Pay).
3. Create an API key on the Dashboard; each key can have its own budget.
4. Point your tool at the right base URL.

API format and tools: one key works in two formats. Anthropic Messages at https://api.apivai.com (no /v1) for Claude Code and the Anthropic SDK; OpenAI Chat Completions and Responses at https://api.apivai.com/v1 for the OpenAI SDK, Cursor, Codex CLI, Cline and Continue. Streaming, tool calling and images work in both formats. Batch API, Files API and embeddings are not offered; the official API has those.

Payment: credit/debit cards, cryptocurrency (USDT and more), Alipay and WeChat Pay.

Pros: lowest per-token price of the five; minutes to set up; Claude and GPT under one key; per-key budgets; every request with its tokens and cost on the Dashboard.

Cons: 60 requests per minute per key by default (higher on request to support); no Batch API, Files API or embeddings; no enterprise contract or cloud-provider invoice.

Fits: individual developers and small teams who want lower Claude Code, Cursor or SDK bills. Full steps are in the Claude Code setup guide and docs.

How pricing works: you buy credits, and each model is generally charged at its provider's list price; a fee applies when you buy credits. See OpenRouter pricing for current terms.

Setup: create an account, buy credits, create a key, and use https://openrouter.ai/api/v1 as the base URL.

API format and tools: OpenAI-compatible, so Cursor, Cline, Continue and the OpenAI SDK connect directly. For Claude Code and other Anthropic-format tools, check OpenRouter's docs for their current guidance.

Payment: cards, plus other methods listed on their site.

Pros: a very wide model catalogue, provider fallback, one bill for many vendors.

Cons: for Claude it is generally not cheaper than going direct.

Fits: people who switch between many model families. We compare it in detail on APIVAI vs OpenRouter.

How pricing works: on-demand Claude prices are generally in line with Anthropic's list prices; batch inference, provisioned throughput and enterprise agreements can change that. Prices can vary by region, so check Bedrock pricing.

Setup:

1. Have an AWS account with billing set up.
2. Enable access to the Claude models in the Bedrock console for your region.
3. Create IAM credentials or a role with Bedrock permissions.
4. Call Bedrock with the AWS SDK or the Anthropic SDK's Bedrock client.

API format and tools: Bedrock uses its own endpoints and AWS request signing, with Bedrock-specific model IDs. The request body follows the Anthropic Messages shape, and Bedrock's Converse API offers a unified alternative. Claude Code can use Bedrock through its own settings (an environment variable plus your AWS credentials); see the Claude Code docs for the current steps. Tools that only accept an OpenAI-style base URL and key need extra work.

Payment: your AWS invoice.

Pros: data stays in your AWS environment; IAM, CloudWatch and AWS contracts and credits apply.

Cons: the most setup of the five; new models can reach Bedrock on a different schedule.

Fits: companies already on AWS that need one invoice and AWS-level controls.

How pricing works: on-demand Claude prices are generally in line with Anthropic's list prices; see Vertex AI pricing for current regional terms.

Setup:

1. Create or pick a Google Cloud project with billing.
2. Enable the Vertex AI API and the Claude models in Model Garden.
3. Check quota for your region.
4. Authenticate with gcloud or a service account.

API format and tools: Vertex AI uses Google Cloud endpoints and auth; the request body follows the Anthropic Messages shape, and the Anthropic SDK has a Vertex client. Claude Code supports Vertex AI through its own settings as well; see the Claude Code docs.

Payment: your Google Cloud billing account.

Pros: fits teams already on Google Cloud; Google Cloud credits and commitments may apply.

Cons: quota requests and region choice add setup; like Bedrock, not a plain base-URL swap for most editors.

Fits: teams whose infrastructure and budget are on Google Cloud.

How pricing works: list prices, with discounts for the right workloads. Cache reads cost a small fraction of normal input, while cache writes cost somewhat more than normal input. The Message Batches API costs about half for jobs that can wait. Details are on Anthropic pricing.

Setup: create an account in the Anthropic Console, add a payment method, buy credits and create a key. Mark the stable part of your prompt (system prompt, documents, codebase) with cache_control.

API format and tools: the reference Messages API, with every official feature: Batches, Files, and new features first. Claude Code and the Anthropic SDK work out of the box.

Payment: card with prepaid credits; invoicing for larger accounts.

Pros: official support and the full feature set; caching cuts repeated input sharply.

Cons: caching does not lower output costs, and output is often most of a coding bill. Caching also works through APIVAI, at APIVAI's cache prices (Claude Sonnet 4.6 cache read $0.12 per 1M tokens).

Fits: teams that need official terms or features such as Batches and Files. See APIVAI vs the Anthropic API.

For most individual developers and small teams, APIVAI gives the lowest price on the same Claude models with the least setup. Bedrock and Vertex AI suit enterprise procurement inside an existing cloud, OpenRouter has the broadest catalogue, and the official API has the full feature set. For a deeper price breakdown, read the Claude API pricing comparison.

Of the five options here, APIVAI has the lowest per-token prices on most Claude models: 56–62% below Anthropic's list as of 2026-10-10. Prompt caching lowers costs further on repeated input.

Generally no. OpenRouter charges around the provider's list price plus a fee on credit purchases; its strength is model variety. Check their pricing page for current terms.

Yes, Claude Code has its own settings for both, using your cloud credentials. Check the Claude Code docs for the current variables and steps.

Yes. Cache reads and writes are billed at APIVAI's cache prices, listed on /pricing.

Yes, the same Claude model IDs are offered; requests pass through to the model providers. APIVAI does not store request or response content, only usage metadata for billing.

Create an APIVAI account and test your own workload from a $10 top-up.