The previous post in this series argued that subscription pricing for AI Coding tools punishes the engineers who use them most. Pay-per-session pricing is the alternative — and the math, when you actually run the numbers, is brutally in favour of the engineer who wants to ship.
This post walks through how Codeforless's pay-per-session model works, what a real session costs, and why an engineer on a budget can outship a team paying ten times more on flat-rate subscriptions.
What pay-per-session actually means
A session on Codeforless is a 1-hour block of dedicated AI Coding capacity. The engineer books the session, gets a fresh browser-based dev environment with Claude Code pre-authenticated and ready to go, and walks away when the work is done. The price is fixed before booking. There is no monthly fee, no rollover, no quota wall mid-session.
The mechanics are deliberately simple:
- Book a 1-hour block. Pay the displayed price. No commitment beyond that block.
- Open the browser to a pre-built dev container. VS Code in the browser, terminal, file browser. Node 20, Python 3.12, Docker, git already installed.
- Claude Code is wired in. Authenticated, configured, ready. Open a terminal, type
claude, start coding. - The session ends, the container is destroyed. Code that should persist gets pushed to your git remote like any other workflow. Nothing personal stays on Codeforless infrastructure.
This is the entire product. Nothing to install, nothing to configure, nothing to clean up.
Time-window pricing — the engineer's discount
Codeforless offers three time-window plans. The same 1-hour session costs different amounts depending on when it runs, because off-peak compute is cheaper for everyone.
| Plan | Window | Price per session (1-pack of 5 sessions) |
|---|---|---|
| Super Hero | Anytime, 24/7 | $19.00 |
| Evening Owl | 16:00 → 08:00 (your timezone) | $14.00 |
| Dracula | 00:00 → 08:00 (your timezone) | $9.00 |
A 20-pack and a 40-pack drop the per-session price further on each plan. The 40-pack of the Dracula plan is the cheapest hour of pre-authed Claude Code on the public internet today.
The point is not that everyone should code at 2am. The point is that engineers who are flexible about when they do their AI Coding work can pay a fraction of the flat-rate subscription price for materially more capacity.
Running the math
Compare a heavy AI Coding user on three different paths:
Path A — Claude Pro at $20/month. Hits the 5-hour rolling window after a couple of focused mornings. Burns the daily limit by Wednesday. Total productive hours per month: maybe 30, maybe 40, with cognitive overhead of constantly batching and pacing.
Path B — Claude Max at $200/month. The weekly quota lifts the ceiling, but the same power user still hits it three days into the week. Total productive hours per month: maybe 60. Cost per actually-productive hour: about $3.30, before counting the days the tool was unusable.
Path C — Codeforless 40-pack on the Dracula plan ($15). Forty hours of pre-authed Claude Code, no quota wall, no shared rate-limiter, fully owned 1-hour blocks. Cost per hour: $0.38. And every hour is a full hour — no warnings, no throttling, no batching anxiety.
Path C ships more code for less money than Path A and dramatically less than Path B. The only reason a heavy user wouldn't take it is if they care more about the convenience of a single subscription bill than they do about the actual cost of the work.
What the engineer trades and what they don't
Pay-per-session is not free of trade-offs. Be honest about them.
The trade. Sessions are time-boxed. If a session expires mid-task, the engineer either books the next one or pauses the work. This is a different rhythm from "I have an open subscription, I can poke at things at any moment." For most engineers actually doing focused work, the time-box is a feature, not a bug — it forces shipping over fiddling.
What stays the same. The model is the model. Claude Code in a Codeforless sandbox is the same Claude Code an engineer would run locally. There is no degraded experience, no different model, no different context window. The work the engineer does in a session is the same work they'd do anywhere else, minus the setup, minus the rate-limit anxiety.
Privacy. Sessions run in isolated containers. Code, prompts and outputs live inside the engineer's environment for the duration of the session and are destroyed when it ends. Codeforless does not retain session contents. The privacy posture is documented at /privacy — short version, nothing about an engineer's session contents is stored or trained on.
Who pay-per-session is for
The model is built for three groups in particular:
- Engineers on a budget. Indie hackers, freelancers, students, engineers between contracts. The flat $200/month subscription is a real chunk of money. The Dracula 10-pack at $15 isn't.
- Engineers with bursty workloads. A side project that needs heavy AI Coding for a weekend, then nothing for a fortnight. Subscriptions are the wrong shape; pay-per-session is exactly the shape.
- Teams trying out AI Coding. Booking a few sessions for a new hire to evaluate the workflow before committing the team to a tool. No procurement conversation, no annual contract.
It is the wrong fit for engineers who want a single tool always-on at the lowest possible per-prompt latency and don't care about cost. The flat-rate subscriptions exist for that audience.
The takeaway
Pay-per-session AI Coding is not a discount product. It is a structurally different deal. The engineer pays for the work they actually do, owns the time-box of each session, and walks away with no commitment when the session ends.
For an engineer on a budget — or an engineer who is just tired of the credit-limit treadmill — the math doesn't favour subscriptions any more. The pricing is on the homepage. A first session is cheap enough to be a Tuesday-night impulse buy.