Anthropic's Startup Program Is Full: What It Gave, and Where Claude Credits Still Are

Anthropic expanded Claude for Startups on 6 October with a free year of Claude Team and $1,000 in API credits. By early November its page said the cohort was full. Here is what still works if you missed it, and what a Claude feature costs without free credits.

Oct 16, 2026

A small startup building with a rocket on its roof receiving a gift box of seats and credit tokens, with a path to a workbench where a product module is being assembled

Claude for Startups is not taking new applications. As of 4 November 2026, Anthropic's program page says: "Our current cohort is now full, and we're not accepting new applications right now." It gives no date for the next cohort; you can subscribe to its newsletter to be told. Current members keep their access. The program had only just been expanded: on 6 October Anthropic added a free year of Claude Team with up to five premium seats, $1,000 in API credits, Claude Marketplace access and office hours with its Applied AI team (TechCrunch). If you missed it, two routes to Claude credits still work, and the real cost of building without them is smaller than most founders assume.


Where Claude credits still are

1. Your VC. Founders backed by a VC in Anthropic's partner network "may be eligible to receive up to $100K in additional API credits", claimed through the VC rather than the application form. Ask your partner whether the fund is on the list. These credits, like the program's, apply only to the first-party Claude API in the Claude Console, not to AWS, Google Cloud or Azure.

2. AWS Activate, spent on Bedrock. Since April 2024, AWS has accepted Activate credits for third-party models on Amazon Bedrock, and Anthropic is one of the providers. So AWS startup credits can pay for Claude, through Bedrock instead of the Console. AWS warns that the post may not reflect current offers, and founders on some regional accounts (AWS India, for one) report trouble subscribing, so check your Activate terms and region before you plan around it. Code-wise it is a small change: Anthropic's SDKs ship a Bedrock client.

3. The next cohort. Subscribe on claude.com/programs/startups. When it reopens, eligibility per TechCrunch's report of the October terms was companies founded in the last five years or funded in the last two.

What the program gave (for members)

  • A free year of Claude Team, up to five premium seats, for the founders' and early team's own work.
  • $1,000 in API credits, for building Claude into the product.
  • Claude Marketplace access, to build and list plug-ins.
  • Office hours with Anthropic's Applied AI team, now run as group sessions every other week.
  • The Claude Startup Stack: offers from AI-native companies building on Claude, which Anthropic values at up to $45,000, redeemed with codes in the Console.

Anthropic's stated reason: "the benefits of AI will reach most people through the companies that build on top of models". Read: they want you building on Claude.

How far $1,000 goes, with or without credits

Take a typical product request, 2,000 tokens in and 300 out, at the October 2026 list prices:

ModelCost per requestRequests for $1,000
Haiku 5.5$0.00035about 2.8 million
Sonnet 5.5$0.007about 140,000
Opus 5.5$0.014about 70,000
Fable 5.1$0.035about 28,000

That is the point founders miss when they chase credits: at the right model, $1,000 is often a year of real usage for an early product, so missing the cohort rarely decides whether you can build. Three decisions decide the bill.

The model per task. When we read an insurer's entire claims book, about 13,000 documents, mostly photos of paper, the whole model bill was under $25, because each step used the cheapest model accurate enough for it (the write-up). In our benchmark of 9 models over 1,260 tests, routing by task cut cost 10 to 100 times. The fastest way to burn $1,000 is to send every request to the largest model.

What you resend. Long system prompts and conversation histories resent on every call add up. Cache what repeats (cache reads cost 10% of input or less) and trim what doesn't.

Whether anyone is watching. Set a usage alert on day one. Then make sure your code stops instead of retrying forever when the balance or a spend limit runs out. We learned that one ourselves: the Claude outage post has the story.

How to spend the first months

  1. Pick one feature users would pay for, not five. The one where AI removes the most manual work.
  2. Test it on real data in week one. Fifty real documents, emails or tickets tell you more than any demo.
  3. Measure accuracy and cost per task before you build the interface.
  4. Design what happens when it is wrong: a review step, an undo, a fallback.
  5. Make it reachable where your users already are. If they live in Claude or ChatGPT, that means an MCP server, and Marketplace access is your distribution.

Where we help

Startups usually don't need an agency for this. They need the first feature built right, then a few hours a week of senior judgment.

  • A $1,900 fixed-price starter: one AI feature or workflow on your own data in about two weeks. You use it for real before deciding anything bigger, and the fee is credited against the build that follows.
  • Fractional CTO time, billed by the minute rather than in day blocks, at €100 to €180 an hour (about $115 to $210): which model, what architecture, what to build first, and what not to build.

Related: Claude API pricing, how much it costs to build an AI agent, fractional CTO services and fractional CTO vs agency.