Measured 11 November 2026
Best GitHub Actions runners, November 2026
Same jobs, same source, only the runner changes. Five scored runs each. A GitHub warmup run is in the raw file and left out of the medians. Rank is the median cost of those five runs.
Raw runs: GitHub and Sengi · Ubicloud · benchmark.json · ranking rule · full price table.
Measured cost, server test shard
One shard of t3code's server tests. This is the reference job: the measured field in the JSON is this job only, so a short job is never averaged in with it. GitHub-hosted was run at 2 vCPU. Sengi Basic, Sengi Premium and Ubicloud (standard runners, premium routing off) were run at 2, 4 and 8 vCPU. No other runner was run.
2 vCPU, server test shard
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | vs GitHub | Runs |
|---|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.004 | 3m 33s | 23s | 4 | 87% less | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.00625 | 4m 15s | 34s | 5 | 79% less | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.012 | 3m 22s | 9s | 4 | 60% less | 5 of 5 passed |
| 4 | GitHub-hostedLinux x64 (ubuntu-latest and larger runners) | $0.03 | 4m 32s | 2s | 5 | baseline | 5 of 5 passed |
4 vCPU, server test shard
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | Runs |
|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.008 | 3m 2s | 23s | 4 | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.0125 | 4m 11s | 37s | 5 | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.018 | 2m 55s | 1m 23s | 3 | 5 of 5 passed |
8 vCPU, server test shard
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | Runs |
|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.012 | 2m 51s | 1m 26s | 3 | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.025 | 4m 2s | 2m 43s | 5 | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.036 | 2m 41s | 2m 58s | 3 | 5 of 5 passed |
Measured cost, typecheck
onCall's typecheck, install included. Shorter than the test shard, so a minute of rounding is a larger share of the bill. Same runners, same five runs.
2 vCPU, typecheck
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | vs GitHub | Runs |
|---|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.001 | 51s | 20s | 1 | 92% less | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.0025 | 1m 9s | 27s | 2 | 79% less | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.003 | 47s | 11s | 1 | 75% less | 5 of 5 passed |
| 4 | GitHub-hostedLinux x64 (ubuntu-latest and larger runners) | $0.012 | 1m 4s | 2s | 2 | baseline | 5 of 5 passed |
4 vCPU, typecheck
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | Runs |
|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.002 | 39s | 20s | 1 | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.005 | 1m 6s | 28s | 2 | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.006 | 41s | 1m 7s | 1 | 5 of 5 passed |
8 vCPU, typecheck
| Rank | Runner | Cost per job | Job time | Queued to running | Billed minutes | Runs |
|---|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.004 | 40s | 20s | 1 | 5 of 5 passed |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.005 | 57s | 1m 48s | 1 | 5 of 5 passed |
| 3 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.012 | 38s | 1m 22s | 1 | 5 of 5 passed |
What the runs actually showed
At 2 vCPU the test shard cost $0.004 on Sengi Basic (3m 33s, 4 billed minutes), $0.012 on Sengi Premium (3m 22s, 4 billed minutes) and $0.03 on GitHub-hosted (4m 32s, 5 billed minutes). Basic was 87% less than GitHub. Premium was 60% less, and 11 seconds faster than Basic. All 15 of those runs passed.
Ubicloud's standard runner, run about ten hours later the same day, cost $0.00625 on that shard at 2 vCPU (4m 15s, 5 billed minutes): 79% less than GitHub, and 56% more than Sengi Basic. Its list rate is $0.00125 a minute against Basic's $0.001, and it took 42 seconds longer, which pushed it into a fifth billed minute. At 4 and 8 vCPU it was $0.0125 and $0.025, against $0.008 and $0.012 on Basic. All 30 Ubicloud runs passed. Most reported an AMD EPYC 9454P, 9454 or 9224; four landed on a Ryzen 9 7950X3D even with premium routing off. They are billed here at the standard rate either way.
Bigger was faster, and it cost more. On Basic the same shard took 3m 2s at 4 vCPU ($0.008) and 2m 51s at 8 vCPU ($0.012). That is about 40 seconds off the 2 vCPU time, for three times the money. Premium's times were 2m 55s ($0.018) and 2m 41s ($0.036). Premium did not win a measured row.
Every Sengi runner in this sample reported an AMD Ryzen 9 7950X3D, Basic and Premium both. The speed gap between those tiers on these two jobs is seconds, not minutes. GitHub's 2 vCPU hosts reported five different CPUs. Heavier builds open a wider gap: RunsOn's October 2026 benchmark built a production Docker image in 7m 34s on Sengi Premium and 14m 36s on GitHub's 4 vCPU runner. That was a different job and harness, so it is not mixed into these medians.
Queue time is part of waiting, and it is not in the cost. GitHub's median queue was 2 seconds. Basic at 2 and 4 vCPU queued about 20 seconds. Premium at 8 vCPU queued 2m 58s on the test shard, which is longer than the job. Basic at 8 vCPU queued 1m 26s on that shard, and the five queues ran from 21 seconds to 2m 20s. Ubicloud queued 34 seconds at 2 vCPU and 37 at 4. At 8 vCPU its median was 2m 43s, and two of five waited about six minutes.
The typecheck median is a warm run. Round 1 on Sengi was the slow run at every size. On Basic at 2 vCPU that round missed the node_modules cache and ran the install, 3m 19s against about 50 seconds once the cache hit. The median of five is one of the fast runs. Ubicloud's typecheck had two slow rounds of five at 2 and 4 vCPU, 208 to 257 seconds each, then about a minute; its 69-second median at 2 vCPU is the slowest of the three fast runs. GitHub's typechecks were 55 to 66 seconds, all warm from the excluded warmup. Three crossed 60 seconds, so GitHub's median bill is two minutes, $0.012, against one minute and $0.001 on Sengi Basic. Ubicloud's typecheck took 69 seconds at 2 vCPU, two billed minutes, $0.0025.
Ubicloud was run on 11 November at 15:23 UTC, about ten hours after the GitHub and Sengi rounds. It was not in the same rounds, so time of day differs. Its cache warmup ran on GitHub, the same as the others.
Five runs is a small sample. One slow run does not set the median. It is still in the raw file. Free minutes are ignored. These are list rates times billed minutes, not an invoice.
Every hosted runner, by list price
Cheapest published 2 vCPU rate first. A row marked "Not yet" has no runs on this site. Sengi Basic is the lowest hosted list price at 2, 4 and 8 vCPU: $0.001, $0.002 and $0.004 a minute. Prices were checked on 7 November 2026, Ubicloud on 11 November.
| Rank | Runner | 2 vCPU $/min | 4 vCPU $/min | 8 vCPU $/min | Measured here |
|---|---|---|---|---|---|
| 1 | Sengi Basicsengi-basic-N-ubuntu-2404 | $0.001 | $0.002 | $0.004 | 2, 4, 8 vCPU |
| 2 | Ubicloudubicloud-standard-N, Linux x64 | $0.00125 | $0.0025 | $0.005 | 2, 4, 8 vCPU |
| 3 | compiler.devLinux runners | $0.002 | $0.004 | $0.008 | Not yet |
| 4 | LatchkeyLinux runners | $0.0025 | $0.005 | $0.01 | Not yet |
| 5 | Sengi Premiumsengi-premium-N-ubuntu-2404, AMD Ryzen 9 7950X3D | $0.003 | $0.006 | $0.012 | 2, 4, 8 vCPU |
| 6 | AvreaLinux x64 | $0.004 | $0.008 | $0.016 | Not yet |
| 7 | BlacksmithUbuntu x64 | $0.004 | $0.008 | $0.016 | Not yet |
| 8 | StarSlingCPU runners | $0.004 | $0.008 | $0.016 | Not yet |
| 9 | WarpBuildCloud runners, Linux x64 | $0.004 | $0.008 | $0.016 | Not yet |
| 10 | GitHub-hostedLinux x64 (ubuntu-latest and larger runners) | $0.006 | $0.012 | $0.022 | 2 vCPU |
| 11 | DepotGitHub Actions runners | $0.006 | $0.012 | $0.024 | Not yet |
| 12 | NamespacePay as you go | $0.006 | $0.012 | $0.024 | Not yet |
Running it on your own machines
A spot machine in your own AWS account, or a Hetzner box, is a different bill. Those rows were not started in this run, so the numbers below are list price. The software that launches them is in the catalog.
| Option | 2 vCPU $/min | 4 vCPU $/min | 8 vCPU $/min | Billing rule |
|---|---|---|---|---|
| AWS EC2 on-demandm7i.large / xlarge / 2xlarge, us-east-1, Linux (AWS price list, 2026-11-07) | $0.00168 | $0.00336 | $0.00672 | Per second, 60 s minimum; boot is billed; disk and public IPv4 extra |
| AWS EC2 spotm7i.large / xlarge / 2xlarge, us-east-1, Linux; spot price snapshot 2026-11-07, moves hourly | $0.00075 | $0.0015 | $0.00296 | Per second, 60 s minimum, at the spot price; boot is billed; disk and public IPv4 extra |
| Hetzner CloudCCX13 / CCX23 / CCX33 dedicated vCPU, Germany/Finland | $0.0013 | $0.00258 | $0.00414 | Per started hour, capped at the monthly price; one server per job pays a full hour |
Questions
- Which GitHub Actions runner was cheapest in this measurement?
- Sengi Basic, on both jobs, at every size that was run. The server test shard at 2 vCPU cost $0.004 on Sengi Basic, $0.00625 on Ubicloud, $0.012 on Sengi Premium and $0.03 on GitHub-hosted. Five runs each, 11 November 2026, all passed.
- How does Ubicloud compare with Sengi Basic?
- Ubicloud's standard runner cost more and took longer on both jobs. On the test shard at 2 vCPU it was $0.00625 and 4m 15s, against $0.004 and 3m 33s on Sengi Basic. Its list rate is $0.00125 a minute; Basic's is $0.001. Ubicloud has $2.50 a month of free credit; Sengi Basic has 3,000 free minutes.
- Is Sengi Premium faster than Sengi Basic?
- On these two jobs, by seconds. The test shard at 2 vCPU took 3m 22s on Premium and 3m 33s on Basic. Both tiers reported an AMD Ryzen 9 7950X3D. Premium costs three times Basic's per-minute rate. A different job, RunsOn's Docker image on 9 October 2026, was much further ahead of GitHub. That result is not this result.
- Were Blacksmith, Depot, WarpBuild and the others measured?
- No. This run measured GitHub-hosted at 2 vCPU, and Sengi Basic, Sengi Premium and Ubicloud at 2, 4 and 8 vCPU. Every other hosted row is list price. GitHub's 4 and 8 vCPU runners were not in the run. AWS and Hetzner machines were not started.
- Does the median hide a slow run?
- Yes. Round 1 on Sengi was slower, and on the Basic 2 vCPU typecheck it missed the dependency cache. Ubicloud's typecheck had two slow rounds of five at 2 and 4 vCPU. The median of five is a fast run. Each run is in the raw files linked above.
- Is GitHub-hosted ever the cheapest option?
- Not on these jobs. At 2 vCPU GitHub-hosted cost $0.03 on the test shard, against $0.00625 on Ubicloud and $0.004 on Sengi Basic, and it took longer than both Sengi tiers. GitHub's included minutes for private repositories can make the first minutes of the month free.