Getting Started

1P vs 3P: Two Ways to Buy Claude, Explained

Same models, two purchasing paths. The right one depends less on the technology and more on how your company already buys, secures, and audits software.

Claude 3P 101 · Updated July 2026 · Unofficial guide

Every conversation about adopting Claude eventually reaches the same fork: do we sign up with Anthropic directly (1P, first-party), or do we consume Claude through the cloud we already run (3P, third-party platforms)? Both paths serve the same models — Opus 4.8, Sonnet 5, Haiku 4.5 — through essentially the same API. The differences are commercial and operational, and they are worth getting right before your first invoice, not after.

What each path actually is

The 1P path is the Claude API at anthropic.com: you create an account with Anthropic, generate an API key, and pay Anthropic directly. Everything Anthropic ships is available there first, and the developer experience is the reference point every other platform is measured against.

The 3P path routes through one of four platforms: Amazon Bedrock (AWS-operated), Google Vertex AI (Google Cloud), Microsoft Foundry (Azure), or Claude Platform on AWS — a distinct fourth option that is operated by Anthropic itself but runs inside AWS, with same-day API parity with 1P. On all of them you use the same official anthropic Python SDK; only the client class, authentication, and (on Bedrock) the model ID prefix change.

The trade-offs, side by side

Dimension1P (Anthropic direct)3P (via your cloud)
BillingNew vendor, direct invoiceFlows through existing cloud bill
Feature timingDay one, alwaysSame-day on Claude Platform on AWS; lags on Bedrock, Vertex AI, and Foundry
Access controlAnthropic API keysYour cloud's native identity (IAM, ADC, resource keys)
List priceReference priceMarketplace list prices match 1P
DiscountsNegotiated with AnthropicCommitted-use terms negotiated with the cloud provider

Notice what is not in that table: model quality. It is the same Claude either way. Teams sometimes assume the cloud versions are quantized, older, or otherwise diminished. They are not — the gap is in surrounding features and their release timing, not the models.

Where the feature gap actually bites

Core functionality — the Messages API, streaming, tool use, vision, extended and adaptive thinking, prompt caching — works on all four 3P platforms. The gaps are in the newer surrounding capabilities: the Batch API, Files API, code execution tool, and web fetch tool are not available on Bedrock or Vertex AI. The web search tool is absent on Bedrock and only a basic variant exists on Vertex AI. Managed Agents exist only on 1P and Claude Platform on AWS. If your roadmap depends on any of these, check the availability matrix before signing anything.

Rule of thumb: If your company has a significant committed-spend agreement with one cloud, start 3P on that cloud — procurement friction usually costs more time than any feature gap. Go 1P (or Claude Platform on AWS) when you need every new capability the day it ships.

The questions that decide it

Who signs? If adding a new vendor takes your procurement team a quarter, 3P via an existing marketplace agreement may be the difference between shipping this year and next. If you can onboard vendors quickly, 1P's simplicity is attractive.

Who audits? Security teams often prefer 3P because Claude usage then appears in the cloud-native logging and identity systems they already monitor, and the deployment largely inherits your cloud provider's compliance posture — confirm the specifics with your provider.

What features, when? Interactive chat and document workloads run fine everywhere. Batch pipelines, file-heavy workflows, and agentic features narrow the field to 1P, Claude Platform on AWS, or (in beta) Foundry.

Is it reversible? Mostly yes. Because the SDK is shared, migrating between 1P and 3P — or between clouds — is primarily an exercise in swapping the client class, credentials, and model ID strings. That portability is a genuine strategic asset: your first choice is not a lock-in decision.

It is also not an exclusive one. A common enterprise pattern is to run production traffic through the 3P platform that matches the company's cloud, while keeping a small direct account for engineers to evaluate newly shipped capabilities before they reach the cloud platforms. The two paths use the same SDK and the same prompts, so nothing built in the evaluation account is throwaway work when the feature later lands where production lives.

Where to go next

If 3P looks right, how to choose between the four platforms is the next read; if you may switch later, see migrating from the Anthropic API to a cloud platform. The feature matrix shows availability at a glance.