cd /news/developer-tools/depot-vs-celesto-ci-what-are-you-pay… · home › topics › developer-tools › article
[ARTICLE · art-170161] src=dev.to ↗ pub= topic=developer-tools verified=true sentiment=· neutral

Depot vs Celesto CI: what are you paying for?

The developer behind Celesto CI published a pricing comparison arguing that Celesto's managed GitHub Actions runners cost roughly 89% less per Linux running minute than Depot's at matched 2 vCPU / 8 GB sizing ($0.000667/min versus $0.006/min), or about nine times the runtime per dollar before allowances. The writeup notes Depot offers a broader platform with larger machines, Windows and macOS support, and more disk, while Celesto offers only Ubuntu 24.04 Linux x64 at 1–8 vCPU with no recurring CI subscription. It stresses the figures are list-rate arithmetic, not a speed benchmark, and that Celesto's public benchmark snapshot does not include Depot.

by read7 min views1 publishedNov 15, 2026

I build Celesto CI. Depot and Celesto both let you keep GitHub Actions while moving jobs to managed runners. The difference is what you want to buy: Depot offers a broader build platform, larger machines and more operating systems. Celesto offers a much lower Linux runtime rate without a monthly CI subscription.

At 2 vCPU and 8 GB RAM, Depot lists $0.006/minute for additional Linux usage. Celesto Memory lists about $0.000667/minute. That is roughly 89% less per running minute, or about nine times the runtime per dollar before allowances. It is not a claim that Celesto finishes builds nine times faster.

This comparison covers Depot's GitHub Actions runners, not its separate Depot CI product. Prices and documentation were checked on November 15, 2026. All prices are USD.

Depot GitHub Actions runners Celesto CI
Linux x64, 2 vCPU / 8 GB $0.006/min additional usage $0.00066667/min, celesto-memory
Linux x64, 4 vCPU / 16 GB $0.012/min additional usage $0.00133333/min, celesto-large
Linux x64, 8 vCPU / 32 GB $0.024/min additional usage $0.00266667/min, celesto-xl
Monthly plan Developer $20; Startup $200; Business custom No recurring CI platform subscription; runtime uses workspace credit
Included usage and trial Developer includes 2,000 base runner minutes/month; Startup 20,000. Seven-day trial advertised One-time $5 Micro trial, up to 20,000 minutes, expires after 30 days; not a recurring allowance
CPU range 2 to 64 vCPU on Linux 1 to 8 vCPU
Platforms Linux x64 and ARM, Windows, macOS Ubuntu 24.04 Linux x64
Local disk at 2 vCPU / 8 GB 100 GB, with a 2 GB memory-backed disk accelerator 40 GB
Dependency cache 25 GB shared cache/registry storage on Developer; 250 GB on Startup; $0.20/GB-month additional storage Default 25 GiB per paid repository and 100 GiB per paid workspace, no separate cache storage charge
Docker Local builds plus a separately priced remote container-build product Docker Engine, Buildx, Compose, service containers and type=gha layer caching
Repository ownership GitHub organization required for this runner product Personal accounts and organizations
Migration Connect Depot, then change runs-on Install the GitHub App, link a funded workspace, then change runs-on

Sources: Depot runner specifications, Depot pricing, Depot runner overview, Celesto setup and sizes, Celesto caching and Docker support.

The CPU and RAM match in the price rows. The machines do not: disk capacity, physical processors, storage and software images differ.

Depot advertises runners up to 3x faster and caching up to 10x faster than GitHub-hosted runners. Those are Depot's claims against GitHub, not measurements against Celesto.

Celesto publishes open workflows, results and source runs. Its November 14 provider snapshot, including a November 15 Docker rerun, does not contain Depot. It cannot settle which of these two providers is faster.

There is still a useful buying question: how much time must a higher-priced runner save to justify its rate?

At the matched 2-vCPU / 8-GB list rates, Depot would need roughly one-ninth of Celesto's billable runtime to reach the same compute charge, before plan fees and allowances. A 2x speedup alone would not offset a roughly 9x rate. That is arithmetic, not a benchmark prediction.

For an actual test, run the same commit several times on both. Keep CPU, RAM and architecture aligned. Separate cold builds from warm-cache runs. Measure queue wait, preparation, dependency restoration, build, tests and cleanup. If a Depot remote builder does some of the work, record its bill too.

The per-minute gap matters most when CI runs all day. Coding agents can increase that volume: more pushes mean more chances to run the same checks.

Here is a Linux-only illustration at 2 vCPU / 8 GB. It assumes equal billable runtime and excludes tax, storage overages, remote builders, discounts, trial credit and billing rounding.

Monthly runtime Celesto Memory Depot Developer, including its 2,000 minutes
2,000 minutes $1.33 $20.00
10,000 minutes $6.67 $68.00
100,000 minutes $66.67 $608.00

Celesto is runtime multiplied by $0.00066667. Depot is $20 plus $0.006 for each minute above 2,000. This is the one-user Developer plan, not a quote for a multi-user team. Depot's Startup plan costs $200 and includes 20,000 base runner minutes, unlimited users and other platform usage. Larger runners consume included minutes through multipliers.

Do not add a plan fee and then charge all the minutes again. Equally, do not ignore the plan fee when evaluating a new subscription.

An existing Depot subscription changes the decision. If you already pay for its other products and have unused included runner minutes, those jobs may add no runner charge. Moving them to Celesto would create new spend rather than remove an existing bill. The table is a new-subscription comparison, not everyone's marginal cost.

Depot tracks usage by the second, with no one-minute minimum, and totals minutes at the end of the billing period. Celesto bills computer runtime, including preparation and cleanup, but not queue time before the computer starts. Compare the actual meters rather than treating the duration of a test command as the whole bill.

Both support caching without replacing compatible actions/cache steps. Celesto also supports setup-action caching such as setup-node and setup-python. Its paid cache allowance is per repository, with a workspace cap; Depot's plan allowance combines cache and registry storage. Those are different units of packaging.

Cache scope differs too. Celesto uses repository and Git-ref permissions, including authorized default-branch restoration. Depot shares a repository's cache namespace across branches, with isolation controlled through cache keys. Check that behavior before migrating workflows that run untrusted contributions.

Depot's remote container builders are a real reason to choose its platform. They run alongside its runners and have their own pricing and included usage. Celesto's runner includes Docker, Buildx and Compose, and can persist build layers through type=gha. That does not make it the same product as a remote builder with a dedicated cache.

Celesto's Docker documentation verifies image builds, Compose, service containers and layer caching. Docker-based actions and whole-job container: configurations need validation for your chosen workflow. Fresh runners also mean the local Docker image store does not persist between jobs.

Depot describes fresh, single-tenant EC2 instances that terminate after a job. Celesto uses single-job JIT runners in fresh microVMs, deleted after the job. Both avoid carrying the previous job's local filesystem into the next one.

That tells you about isolation, not long-run availability. There is no head-to-head reliability study in the evidence used here.

Depot advertises unlimited concurrency, but explicitly warns that macOS capacity is not fully elastic and can queue during high demand. Celesto concurrency depends on workspace limits, credit and available capacity. Test the burst of jobs your repository actually produces, not only one successful run.

Install the Celesto CI GitHub App, select your repositories and finish workspace setup. Use eligible paid or promotional credit for Memory or Large. The $5 trial is Micro-only.

For a Depot 2-vCPU / 8-GB job:

 jobs:
   test:
-    runs-on: depot-ubuntu-24.04
+    runs-on: celesto-memory

For a 4-vCPU / 16-GB job:

-    runs-on: depot-ubuntu-24.04-4
+    runs-on: celesto-large

Do not substitute bare celesto for a size-matched test: it selects Micro, with 1 vCPU and 2 GB RAM.

Keep compatible steps, but check tool versions. Celesto's image is not GitHub's hosted image. Caches live in a separate service, so expect the first run to start cold. Move one job first; GitHub lets different jobs in the same workflow use different runners.

Depot is worth paying for when its broader platform or a faster complete workflow earns the premium. Celesto is the stronger starting point when the requirement is straightforward: keep GitHub Actions, run Linux checks, and reduce recurring compute spend.

The published rates put Celesto about 89% below Depot's additional Linux runtime rate at matched CPU and RAM. That is a substantial margin without needing to claim an unmeasured speed win. Test one representative workflow, include the allowances you actually use, and compare the bill alongside the wait for a green check.

Per Linux runtime minute at the matched sizes above, yes: roughly 89% less. Your total depends on runtime, Depot's plan and included minutes, other platform usage, storage and discounts. Unused minutes in a subscription you already pay for can change the marginal-cost answer.

This article does not establish that. The cited Celesto provider snapshot does not include Depot, and Depot's GitHub comparisons do not answer a Celesto comparison.

Yes. Select runners per job. Keep platform-specific or remote-builder-dependent jobs on Depot while testing compatible Linux jobs on Celesto. Any retained Depot subscription or remote-builder usage still has its own cost.

No. The one-time $5 trial covers Micro, expires after 30 days and is shared with sandbox usage. Memory and larger sizes need eligible workspace credit.

Get started: Try Celesto CI

── more in #developer-tools 4 stories · sorted by recency
── more on @celesto ci 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/depot-vs-celesto-ci-…] indexed:0 read:7min 2026-11-15 · —