Two of October’s useful development routes let you postpone signup: a temporary Postgres project and a Linux machine reached over SSH. The database lasts three days; the machine gives you an hour to build. The claim links and account handoffs have their own limits.
This is the 1 October 2026 check, based on official pricing and documentation. No accounts or workloads were created for it. The maintained free-compute guide carries the broader shortlist, and the September edition is the previous snapshot. The focus here is getting a prototype started, then keeping its expiry and credit conditions attached to it.
A database before its owner signs up
An agent can create a Claimable Neon Postgres project before a human has an account. Neon announced that route on 11 September.
The unclaimed project gets 100 MB storage and 1 GB transfer. There are two deadlines:
- The project expires after 72 hours.
- Its claim code expires after 15 minutes.
A database can still exist after its original claim code has expired. Treating those clocks as one deadline would miss the shorter one.
The Data API and Auth can be requested at creation. Functions, Object Storage and AI Gateway require claiming first. That leaves a useful temporary route for generating and testing a database-backed prototype, with a clear boundary around which backend products are available before onboarding. It does not replace the permanent Free plan or a backup.
A shell for the development session
Railway’s equivalent entry point is a command:
ssh railway.new
The 25 September announcement describes a Linux VM with 2 vCPU and 2 GB RAM, coding agents already installed, and a private preview URL. It needs neither an account nor a payment card.
Here the active development window is 60 minutes, followed by 24 hours to claim the VM. Claiming transfers the machine, files and URL into an account and makes the preview shareable. Unclaimed machines and files are deleted. Save or claim work before the clock runs out.
The trial documentation adds three anonymous boxes per IP address per day and regional capacity restrictions. Claiming also changes the billing context: the workload moves into the account’s ordinary credits and billing. The separate signup trial is $5 for 30 days, followed by $1 of recurring monthly Free credit. An anonymous hour and an account’s monthly allowance are different offers.
After claiming, read the backend meters separately
Neon’s backend became generally available on 18 September. Functions and Object Storage are available in Ohio, Northern Virginia, Frankfurt and Singapore, replacing September’s Ohio-only beta description with published Free allowances.
The Free pricing page lists up to 100 Postgres projects, each with 100 CU-hours per month and 0.5 GB database storage. Object Storage gets 5 GB per project. Functions get 10 active Capacity-Hours, 400 waiting Capacity-Hours and one million monthly invocations; Auth includes 60,000 monthly active users.
Those figures cannot all be multiplied by 100. The plan documentation applies the Functions limits across the account as well as independently per project. A project’s 5 GB network allowance is shared by Postgres, Object Storage and Functions.
Active Capacity-Hours measure CPU work; waiting Capacity-Hours measure time waiting on I/O. Adding them together does not produce 410 hours of an arbitrary always-on server. AI Gateway also uses purchased prepaid credits, so the backend launch does not supply free model inference.
A compute credit may exclude the API you call
For a GPU or model experiment, the product covered by the credit matters as much as its size.
Modal’s $30 is recurring compute credit. Its current pricing FAQ covers CPU, GPU and memory usage in Functions, Sandboxes and Notebooks. Shared Endpoints tokens are excluded and billed from the first request, effective 1 September 2026. The Workspace spend limit is the cash cap; a usage budget measures usage before credits. A payment method is required, and extra compute can charge.
Lightning’s GPU credits are a starter balance. The current offer gives five credits on registration and 25 more after adding a payment card. Unused credits expire after 12 months. GPU work spends that balance. One CPU Studio can remain free with a restart every four hours. September’s claim of 15 recurring monthly credits is not supported by today’s offer; no dated announcement was found to establish when it changed.
Mistral publishes a monthly API allowance. Its Free pricing now specifies $10 per month in API credits. The subscription documentation shares usage across Studio, API and Vibe Code. With pay-as-you-go disabled, exhausted usage can stop until renewal; enabling it allows additional tokens to bill. Model access and rate limits still need checking for the account. This is a current pricing observation, with no established launch date.
Batch work can wait; VM promotions can end
A prototype’s non-urgent processing has another dated option. Cloud Run added delayed job execution in Preview on 8 September: provisioning can wait up to 12 hours for a lower price, and execution can then run for up to 12 hours.
The pricing table gives a monthly free equivalent of 342,857 vCPU-seconds and 642,857 GiB-seconds at Tier 1 pricing. Google applies the free tier as a spending discount aggregated by billing account. The delayed-job, ordinary-job and request-based rows are not independent grants to add together. A billing-enabled account is involved, and builds, stored images, networking and excess usage can charge. The scheduling documentation carries the execution limits.
For a longer-running ARM VM, the AWS T4g promotion still provides 750 aggregate t4g.small hours per month, ending 31 December 2026. Disks, public IPv4 and surplus CPU can charge. Its end date needs to travel with the instance-hours figure.
The temporary database and SSH machine give a prototype somewhere to start. Their short claim windows remain relevant even while the resources still exist. Once work moves into an account, the covered product, shared meters and billing settings determine what the headline allowance actually buys.
Want to talk about something I’ve written or built? Get in touch.
This article was adapted with AI assistance from an original article on billiem.uk. The original article was reviewed before publication.
Top comments (0)