Claude Code's free credits only spend in the cloud

Free credits are never just free credits. Anthropic is giving Pro and Max subscribers up to $250 in Claude Code cloud session credits, and the interesting detail is the restriction: they do not spend on your own machine. This is a migration incentive with a six-week fuse, and it is worth taking — deliberately, with a decision about where your repo runs.

What actually happened

Per BleepingComputer, the promotion started September 23, 2026. Pro subscribers get $100 in promotional credits; Max subscribers get $250. Eligibility is limited to individual Pro and Max subscribers with an active subscription when the promotion began.

The dates are the part to write down. Claim by October 7. Credits expire November 4. That is roughly four weeks of useful life after the claim deadline.

Credits apply only to cloud sessions — agent runs on Anthropic-hosted infrastructure, started from claude.ai, the mobile or desktop apps, or the CLI with claude --cloud. Local sessions on your own computer do not draw from them. The genuinely useful mechanic: when you hit your plan's local usage limits, cloud sessions keep running on promotional credit. Once the credits are gone, cloud sessions fall back to standard plan usage limits, with no separate container charge.

Why cloud agent sessions matter for your business

Claim it, but know what moves. A cloud session means your repository is checked out on someone else's machine, and the agent's shell runs there. For an open-source project or a greenfield prototype, that is a non-issue and a genuine speed win. For a client codebase with production credentials in the environment, that is a decision your contracts may already have an opinion about. Read the data-handling terms once, then apply the answer consistently.

The real product here is unmetered-feeling parallelism. The pitch is not "cheaper tokens," it is that local rate limits stop being the thing that ends your afternoon. Cloud sessions run detached, so three agents can work three branches while you do something else. If your workflow is currently one agent and a lot of waiting, that is the habit the credits are trying to change.

Test what you would actually keep paying for. Four weeks of credit is a real evaluation window, so spend it on your slow, annoying work — migrations, dependency bumps across a monorepo, backfilling tests — not on demos. At the end you will know your own cost per completed task, which is the only number that matters when the credit runs out on November 4.

Keep the local path working. We use cloud agents where they fit and keep everything reproducible on a laptop, because a workflow that only functions on one vendor's hosted runtime is a dependency you did not price. Same repo, same scripts, same test command, either place. Promotional pricing is how a runtime becomes load-bearing before anyone decided it should be.

Sources: BleepingComputer, Claude Code documentation.