What will your E2B workload cost each month?

Price aggregate running seconds using your E2B runtime quote, then add the published plan fee and any concurrency expansion. Replay the busiest workload to see which fee threshold it crosses.

Find the burst that adds $500 a month

E2B's published Pro expansion comes in blocks of 500 concurrent sandbox slots at $500 per month each. Compare actual start and stop timestamps for your busiest repeatable batch. If the trace tops out at 101 concurrent sandboxes, the first add-on buys 500 slots for a one-slot overage. Run a second trace with a dispatcher ceiling of 100 and measure the last job's completion time against the batch deadline. For example, an illustrative 101-job burst with identical 40-second jobs and no startup overhead finishes in 80 seconds under that ceiling, versus 40 seconds with enough slots. The extra 40 seconds may be acceptable for a nightly batch and unacceptable for an interactive workload. Recompute using the measured duration of the job that actually waits.

How do I increase my concurrency limit? - E2B Docs

Reconcile running seconds with the sandbox ledger

E2B documents that paused sandboxes consume neither running concurrency nor compute charges. Export timestamps for creation, pause, resume, and stop, and expand each lifecycle into active intervals. Join them to the job ledger so chargeable retries can be separated from user-visible successes. For example, a sandbox active for 35 seconds, paused for 10 minutes, and active for another 15 seconds contributes 50 running seconds. Its retry on a second sandbox contributes separately even if the first attempt produced no usable result. Sum active seconds independently of your request count and compare the total with the same billing window. If totals diverge, inspect resume events and jobs that returned a result before the sandbox stopped. Use a representative mix of successful and failed batches for the monthly projection.

Do paused sandboxes count toward the concurrency limit? - E2B Docs

Measure the busiest minute first

E2B's published Hobby plan allows sessions up to one hour and 10 concurrent sandboxes. Pro allows sessions up to 24 hours and includes 100 concurrent sandboxes. Enterprise lists multi-day and persistent sessions with custom billing. Measure the longest uninterrupted run and the busiest interval using sandbox start and stop timestamps. Include retries and overlapping attempts in that record.

In an illustrative burst, 14 independent jobs arrive together and each runs for four minutes. With 10 slots, the first 10 can start immediately and the last four wait roughly four minutes, assuming no startup overhead. Keep Hobby in contention if that delay meets your deadline and the required sandbox shape is available. If the jobs must all start immediately, price Pro.

Pricing | E2B โ€” The AI Agent Cloud

Budget running seconds and retries

E2B charges by the second for a running sandbox. For each template, record allocated vCPUs and RAM, then enter the shape and aggregate running hours in E2B's pricing estimator. For a rate-card calculation, verify the CPU and memory rates and their units, multiply each by its allocation and the running seconds, and sum across shapes. Check whether storage, credits, or other charges affect the estimate in your billing console.

For illustration, assume a sandbox shape costs $0.18 per running hour. Two hundred accepted jobs plus 20 full retries at six minutes per attempt produce 22 running hours and $3.96 in compute. On Pro, that gives $153.96 before add-ons, storage, model usage, or taxes. Reconcile a small pilot with billed usage before extrapolating.

Pricing | E2B โ€” The AI Agent Cloud

Check for accidental wake-ups

Paused E2B sandboxes consume neither a running concurrency slot nor compute charges. If a job can stop between phases, pause it when the first phase finishes and resume it when the next input is ready. For auto-resume, check whether health probes, SDK reads, or other routine traffic can wake the sandbox. Record lifecycle transitions alongside the job's own timestamps.

In an illustrative workflow, a sandbox does three minutes of work, waits 40 minutes for a reviewer, then runs two final minutes. The target is about five running minutes plus lifecycle overhead. If the trace shows the sandbox running through most of the review gap, look for a missed pause or an unwanted wake-up. Move status checks outside the sandbox and rerun the same sequence.

Do paused sandboxes count toward the concurrency limit? - E2B Docs

Put a price on starting sooner

E2B's self-serve Pro concurrency add-on costs $500 per month for 500 extra slots, on top of usage. Pro can reach 1,100 concurrent sandboxes through add-ons. Before buying a block, test a queue capped at the included limit and measure arrival-to-start delay, completion time, and failed or retried jobs. Compare those results with a realistic deadline and confirm the add-on in the console billing tab.

For an illustrative case with 130 simultaneous five-minute jobs, 100 slots leave 30 jobs waiting roughly five minutes if durations are uniform. If that delay is acceptable, keep the queue. If the jobs repeatedly miss a firm deadline and the extra capacity resolves it, budget $650 per month for Pro plus one block, with usage charged separately.

How do I increase my concurrency limit? - E2B Docs

Sources