Linear coding sessions pricing: how AI credits, tokens and sandbox time add up

What changed between June and October 2026

The 20 August change is the one that matters for budgets. Linear calls it "a lower-cost pricing model with two components". Its changelog does not say what the earlier model was, so we cannot give a before-and-after comparison.

How Linear AI credits work

The rules come from Linear's AI credits documentation:

• One pooled balance. Credits are held in US dollars at the workspace level and shared across every AI feature that uses them.
• Opt-in. A workspace that never adds funds is never charged. Coding sessions and Loops simply cannot start without a balance.
• Who can pay. Only workspace admins (owners on Enterprise) can add credits, by card only, through Stripe, on an invoice separate from the subscription.
• Top-ups. The minimum manual top-up is $10. Automatic Reload tops up when the balance falls below a threshold you set; its minimum is $50.
• Expiry. Purchased credits expire 12 months after purchase. Unused balance rolls over between billing cycles until then. Expired balances cannot be refunded or transferred.
• Order of use. Promotional credits are spent first, then credits issued by Linear support, then credits you bought, oldest-expiring first within each group.

What does not use credits matters as much. Linear's documentation says "other Linear AI features are included in your plan and do not draw from your AI credits balance". Linear Agent itself is listed on every plan from Free. Only coding sessions, where the agent writes code and opens pull requests, and Loops, where it runs agent work in the background, are metered.

Plan access differs between the two. Linear's coding sessions documentation now lists Free, Basic, Business and Enterprise; the June launch post said Basic and above. Loops are on paid plans only.

What a single coding session costs

A coding session's cost has two parts, and only one of them has a fixed price.

Sandbox runtime is $0.25 per 20-minute block. Linear prices it per block, not per minute, so assume a session that runs 25 minutes is billed for two blocks. A one-hour session costs $0.75 in runtime.

Model tokens are charged at whatever the provider publishes for the model you chose, with no markup. This is the variable part. It depends on the model (Linear's documentation lists Claude, GPT and GLM models, with GPT-6 Sol as the current default under Auto), on how much of your codebase the agent reads, and on how many follow-up prompts you send. Adaptive routing, added on 24 September, sends simpler tasks to a faster model, which should lower the average.

Linear does not publish a typical per-session figure. The usage dashboard shows the breakdown by model after the fact. So the honest answer to "what does a session cost?" is: $0.25 to $0.75 for runtime, plus a token cost you will only know once you have run a few.

What Loops cost

Loops are recurring agent workflows triggered by a schedule, by issue, project or cycle events, by GitHub pull request events, or by messages in Slack channels. Linear's AI credits page says "typical runs without a coding session cost $0.07–$0.20". A Loop that starts a coding session also pays that session's tokens and runtime.

A daily Loop that never writes code costs roughly $2 to $6 a month at those rates. Ten such Loops across a team cost $21 to $60. Event-triggered Loops are harder to predict, because the number of runs follows the number of events: a Loop on every new triage issue runs as often as issues arrive.

What this means for a 5-person team's bill

The seat bill is fixed: 5 people on Basic cost $50 a month billed yearly, and on Business $80. The AI bill moves with use. To show the range, we picked two illustrative token costs per session, $0.50 and $3. These are assumptions, not Linear figures; your dashboard will show your real number. Each session is assumed to run two 20-minute blocks ($0.50 runtime).

Add three daily Loops and you add about $6 to $18. The pattern is clear even with made-up token figures: once a team uses coding sessions regularly, the AI line can be larger than the seat line, and it varies month to month with how much code the agent writes.

That is not a criticism. Paying provider rates with no markup is a fair deal, and you only pay for work the agent actually did. It does mean the total bill stops being a number you know in advance.

How to keep a Linear AI bill predictable

1. Turn off Automatic Reload at first. Buy a fixed top-up, such as $50 or $100, and watch how long it lasts.
2. Set a workspace spend limit in Settings, Usage and limits, with a monthly reset. When a limit is reached, new usage pauses until it resets or an admin raises it.
3. Add per-user limits for the people who run the most sessions, and per-Loop limits for event-triggered Loops.
4. Pick the model deliberately. Token rates differ widely between models. Leave Auto on only if you are happy with its choice.
5. Review the usage explorer monthly and move the spend limit to match what the work is worth to you.
6. Track expiry. Bought credits last 12 months; promotional credits have their own dates in the dashboard.

Included runs instead of metered credits: how Atrix differs

We make Atrix AI, which takes the other approach, so read this section knowing that. AI runs are part of the plan: Builder is $39 a month for 5 seats and 100 runs; Studio is $149 a month for 15 seats and 500 runs; yearly billing is 20% less. Runs reset each month. When they are used up, the agent tells you so and waits for the next month or an upgrade; there is no overage charge, so the bill is the plan price. Chatting with the agent does not use a run.

The trade-off is real. A Develop run turns a brief into requirements, an architecture note and planned issues on the board; a Market run works on go-to-market. The AI agent does not write code, open pull requests or run code in a sandbox, which is exactly what Linear's coding sessions do. Teams that want code written connect Claude Code, Codex or Cursor through the MCP server, and those tools bill on their own subscriptions. If your main need is an agent that ships pull requests inside the tracker, Linear is the stronger product.

The Linear comparison sets out the rest, and Linear alternatives in 2026 covers other trackers.

FAQ

How much does a Linear coding session cost?

Since 20 August 2026, a coding session costs the model's tokens at the provider's published rate, with no markup, plus $0.25 for each 20-minute block of sandbox runtime. Linear does not publish a typical total; the usage dashboard shows each session's cost.

Are Linear AI credits included in the subscription?

No. AI credits are a separate, prepaid workspace balance bought by card, billed on its own invoice. Linear has given promotional credits at launches, such as $20 per seat when Loops launched on 20 July 2026. Other Linear AI features, including Linear Agent outside coding sessions and Loops, are included in the plan.

How much do Linear Loops cost?

Linear says typical Loop runs without a coding session cost $0.07 to $0.20. A Loop that starts a coding session also pays for that session's tokens and runtime.

Do Linear AI credits expire?

Yes. Purchased credits expire 12 months after the purchase date. Promotional credits have their own expiry dates, shown in the usage dashboard. Unused credits roll over between billing cycles until they expire.

Can I cap Linear AI spending?

Yes. Admins can set spend limits for the workspace, for each user and, since 8 October 2026, for each Loop, with daily, weekly or monthly resets. When a limit is reached, new AI work pauses until the limit resets or is raised.

Next step

Compare the two models side by side on the Linear comparison, or see what a fixed number of runs looks like on Atrix pricing.