Reason, the cloud coding agent from Reason Machines, sells two self-serve plans: Pro at $50 a month and Max at $200 a month, each with a weekly allowance of AI usage and cloud-computer time, and both let you connect your own models. This breakdown explains what the credits actually buy, when bringing your own keys makes the seat pay for itself, and how the vendor's "22–30% lower cost per task" claim holds up against its own published numbers. We also put it next to Codex and Cursor plans so you can see which bill fits your work.
What Reason is, in one paragraph
Reason is an agent harness: the layer that decides which tools a model calls, how often it goes back to the model, and where the work runs. Sessions run on cloud computers (cloud Linux, a cloud Mac, your own Mac or an SSH box), you can start them from a web app, an API or by mentioning @Reason in a Slack thread, and the result comes back as a pull request, a document or an answer. Its November 9, 2026 launch post says the product reached 20,000 users in its first four weeks. The pitch is not a new model. It is that the same model costs less when the harness wastes fewer tokens, and that you can run it on model subscriptions or API keys you already pay for.
Reason plans and prices (checked today)
All prices are in US dollars, per person, from Reason's own "Plans, prices and credits" documentation page.
- Pro, $50 a month per person. The frontier Reason model, bring-your-own models, cloud computers, inviting teammates, and a weekly allowance of $50 of AI usage plus $50 of Cloud usage (1,250 credits each). Eligible new customers get a 3-day free trial; a card is required and the plan renews at $50 unless cancelled in Settings › Billing.
- Max, $200 a month per person. Everything in Pro with four times the weekly allowance ($200 AI + $200 Cloud, 5,000 credits each), early access to new models and features, and 24/7 support. No trial.
- Enterprise, contact sales. Custom weekly limits per seat plus a shared pool, private API and OIDC auth, SOC 2 and HIPAA, on-prem workers and custom capacity.
There is no longer a free plan for new users. Workspaces that already had "Free with ads" keep it, but it only covers your own messages in the Reason app on the default model, and excludes Loops, automations and anything started from the API, CLI or MCP. Earlier $20 Pro and $75 Team plans are also grandfathered for existing customers only. YC companies can claim $1,000 in credits through a Bookface deal.
How the credits and weekly allowances work
One credit = $0.04
Reason measures usage in credits, and one credit is four cents, so $1 equals 25 credits. That is why Pro's "$50 of AI usage" is written as 1,250 credits. Two separate allowances draw on those credits:
- AI usage covers model calls that Reason pays for, plus images, voice and transcription.
- Cloud usage covers the time cloud computers are running.
Both reset every week. The app shows them as percentages, not dollar amounts, in the usage popover and in Settings › Billing. If you run out of one, only that kind of work stops: cloud computers pause when Cloud usage is gone, and model turns are refused when AI usage is gone, until the reset, a top-up, or until you switch to a model you connected yourself.
The monthly math nobody prints
Because the allowance is weekly but the bill is monthly, the headline understates what Pro can include. A year has 52 weeks, so the weekly $50 becomes about $216.67 a month of AI usage (50 × 52 ÷ 12) and the same again for cloud time: roughly $433 of metered allowance for a $50 seat, if you actually use every week. Max scales the same way to about $867 of AI usage and $867 of cloud time a month for $200. The catch is the word "if": the documentation says both allowances reset weekly and does not mention any rollover, so a quiet week is simply lost value. Teams that work in bursts should budget on what they use in a typical week, not on the theoretical monthly total.
Top-ups
When a week's allowance runs out you can buy a one-time top-up of $20, $50, $100, $200 or $500. Top-up credits never expire, are shared across the workspace, and are drawn only after a person's own weekly allowance is used. Promotional and YC credit and an Enterprise shared pool behave the same way.
Bring your own keys: when the seat pays for itself
This is the part that makes Reason different from most coding-agent plans. Pro and Max both include bring-your-own models: ChatGPT, Claude, GitHub Copilot and Grok subscriptions, API keys, gateways and custom endpoints. Models you connect yourself are billed by that provider and count toward neither allowance. Your Reason seat then mostly pays for the harness and the $50 a week of cloud-computer time, and the model bill stays wherever it already is.
So when does $50 a month make sense purely on token savings? Here is a simple break-even, using the vendor's own efficiency numbers (covered in the next section), not our own tests:
- If Reason really saves 22% per solved task, the seat covers itself once the same work would cost you about $227 a month in API spend elsewhere ($50 ÷ 0.22).
- At the low end of the vendor's per-attempt range, 15%, the break-even rises to about $333 a month ($50 ÷ 0.15).
- Below roughly $200 a month of model spend, token savings alone will not cover the seat. You would be paying for the cloud computers, Slack workflow and parallel sessions instead, which may still be worth it, but it is a different decision.
If you connect a flat-rate subscription such as ChatGPT Plus or Pro instead of an API key, the saving does not show up on a bill at all. Fewer tokens per task means the subscription's usage limits last longer. Reason's provider docs also note that "Fast mode" on a ChatGPT subscription asks for the provider's priority tier, which answers sooner but uses the plan faster, so leave it off for long background jobs.
Checking the "22–30% cheaper per task" claim
Reason makes two related claims. The launch post says "up to 30% lower cost per task than Codex, Cursor or Devin." The more detailed Reason 1.1 write-up (October 10, 2026) gives specific figures on GPT-6 Astra at xhigh effort:
- Coding Agent Index v1.5: Reason 65.1 vs Codex 62.5, with Pi at 61.4.
- Cost per solved task: 22% below Codex and 10% below Pi.
- Cost per attempt: 15–23% less than Codex on every benchmark in the index.
- Tokens per attempt: 1.92M input tokens vs 2.48M for Codex, and 15% fewer output tokens.
- On GPT-6.1 Sol, a separate chart reports a 34% lower median cost per task against Codex at the same effort.
What we could check: the token figures are internally consistent with the headline. 1.92M is 22.6% fewer input tokens than 2.48M, and input (mostly cached input) dominates agent bills, so a ~22% lower cost per attempt is what you would expect. The write-up also explains where the saving comes from: a "codemode" tool that lets the model write one JavaScript program to run several tools in parallel (their example finishes in 3 model calls instead of 7), a small fixed tool set, and a prompt that never changes between turns so every request stays cache-eligible.
What we could not check: these are the vendor's own reproductions, and the post itself says they are not endorsed by Artificial Analysis. The runs, methodology and scorer are public on GitHub and Harbor, which is better than most launch benchmarks, but they cover one model family on one benchmark suite. "Up to 30%" is an upper bound; the 22% figure is the one to plan with, and your own repository may land higher or lower. We have not run Reason ourselves for this post, so treat every saving here as the vendor's number, not ours.
Reason vs Codex vs Cursor: plans side by side
The comparison below uses only each vendor's official pricing page, read today. "Included usage" is not comparable unit for unit, because each product meters differently.
| Plan | Price | Included usage | Own models / keys | Cloud agents |
|---|---|---|---|---|
| Reason Pro | $50/mo per person (3-day trial, card required) | $50 AI + $50 Cloud usage per week (1,250 credits each) | Yes: ChatGPT, Claude, Copilot, Grok subscriptions, API keys, gateways; billed by provider, outside the allowance | Yes, cloud computers + Slack |
| Reason Max | $200/mo per person | $200 AI + $200 Cloud usage per week | Same as Pro | Yes, plus 24/7 support |
| Codex (ChatGPT Plus) | $20/mo (Free $0 and Go $8 tiers are desktop-app only with GPT-6 Luna) | Shared ChatGPT/Codex limits; e.g. 5–45 GPT-6 Astra local messages per 5 hours; extra credits purchasable | Separate API-key route: pay API prices, CLI/SDK/IDE only, no cloud features | Yes: cloud tasks, code review, Slack integration |
| Codex (ChatGPT Pro) | $100, $200 or $500/mo | No five-hour limit currently; weekly limits may apply | Same API-key route | Yes |
| Cursor Premium | $30/mo (Free $0; Super $100; Ultra $200) | Shared AI usage pool, 4x Plus; third-party models at API list price, billed separately | Third-party models billed by usage | Yes: cloud agents & automations, Bugbot reviews |
The pattern is clear once it is laid out. Codex and Cursor are cheaper to start and bundle their own model usage. Reason costs more per seat but is the only one of the three whose plan is designed around running other providers' subscriptions and keys outside its own meter, with cloud time included.
Which plan fits which team
Pick Reason Pro if
- You already pay a few hundred dollars a month in model API spend for agent work and want that work to use fewer tokens.
- You want background agents started from Slack or an API on cloud machines, and you would otherwise pay separately for sandboxes.
- You have several model subscriptions and want automatic failover between them when one hits a limit.
Stay on Codex or Cursor if
- You code interactively in an editor most of the day and your $20–$30 plan rarely runs out.
- Your team is small and spending under about $200 a month on models: the break-even above will not be reached on tokens alone.
Consider Max only if
- One person regularly exhausts Pro's weekly allowance before the reset. Top-ups ($20 and up, never expiring) are a cheaper bridge for occasional heavy weeks.
Editor's note
Who it's for: teams running many parallel or long-running agent jobs, and people with existing model subscriptions who want one harness on top. Not for: hobbyists looking for a free tier; there isn't one for new users.
Gotchas we found in the docs:
- Every member needs a seat. In a workspace with other members, each person needs their own Pro or Max seat or they only see the paywall. Inviting from Settings › Team & Members charges the workspace card for each Pro seat when the person joins; an invite sent through the API does not pay a seat.
- "Model turns refused" usually means the week's AI usage is used up. Fixes: wait for the weekly reset shown in Settings › Billing, add a top-up, or switch the session to a model you connected yourself.
- Cloud computers paused means Cloud usage ran out, even if AI usage is left. The two allowances are tracked separately.
- iPhone purchases are different. A plan bought in the iPhone app (and a trial) has one shared pool instead of two allowances, and is managed by Apple.
- Fast mode drains subscriptions faster. Leave it off for overnight or batch jobs.
FAQ
Is Reason free?
Not for new users. The old "Free with ads" plan is kept only by workspaces that already had it. New customers can try Pro for 3 days with a card, and YC companies can claim $1,000 in credits.
If I connect my own API key, do I still use up my weekly credits?
No. According to Reason's plans page, models you connect yourself are billed by that provider and count toward neither the AI usage nor the Cloud usage allowance. Cloud computer time still draws on Cloud usage.
Is Reason really 30% cheaper than Codex?
"Up to 30%" is the launch headline. The detailed write-up reports 22% lower cost per solved task and 15–23% lower cost per attempt on GPT-6 Astra, from the vendor's own reproduction that Artificial Analysis has not endorsed. The token counts it publishes are consistent with that figure, but your results will depend on your codebase.
Related reading on lightmind24
- Codex CLI Not Working? Fixes for Login, Sandbox and Limits
- OpenRouter Pricing: Free Models, 5.5% Fee and BYOK
- GitHub Copilot Pricing: Free, Pro, Pro+ and Max Credits
Bottom line: Reason's $50 seat is a bet on token efficiency, so do the break-even with your real monthly model bill before you switch.
Screenshots of Reason's plans documentation and launch post, plus our own comparison table, are included at the end of this post. Image credit: screenshots from reasonmachines.com and docs.reasonmachines.com; comparison table by lightmind24.
lightmind24 is an independent review site, not affiliated with Reason Machines, OpenAI or Cursor.
Last verified: 2026-11-15
Sources:
Reason: Plans, prices and credits
Reason: Provider accounts and Fast mode
Reason launch post (Nov 9, 2026)
Introducing Reason 1.1 (benchmark write-up)
Reason for Startups (YC credits)
Codex pricing
Cursor pricing
댓글
댓글 쓰기