I really appreciate GitHub taking this feedback seriously and reconsidering the pricing model. I'd like to share a perspective that I think represents a huge part of your community: hobbyists tinkering in home labs, junior DevOps engineers building their skills, and teams operating on tight budgets. For all of us, self-hosted runners aren't just nice to have, they're essential, both from a security perspective and a financial reality. And let's be honest, when you're learning this stuff, your workflows tend to run longer as you experiment and iterate. A per-minute model hits that learning phase especially hard.
The Security Reality
For anyone running isolated internal services like HashiCorp Vault, S3-compatible storage systems, or other sensitive infrastructure, self-hosted runners aren't optional, they're mandatory. The alternative would be punching massive security holes in your firewall to allow GitHub's infrastructure and potentially millions of users direct access to your internal systems. That's just not a viable option from a security standpoint.
Many organisations have architectures where critical automation must use self-hosted runners. We shouldn't be penalised for choosing security over convenience.
Why Per-Minute Billing Doesn't Work
The $0.002/minute charge creates real problems for teams trying to manage their infrastructure responsibly.
First, you can't forecast costs reliably. When billing is tied to minute-level granularity, organisations can't easily calculate what their automation will cost. A workflow optimisation that shaves 30 seconds compounds across thousands of runs monthly, but there's no way to predict the actual impact on your bill.
Second, the billing model actually punishes efficiency. If GitHub bills in 1-minute increments, a 15-second job costs the same as a 60-second job. That incentivises people to just let workflows run longer rather than spending time optimising them, which seems backwards.
Third, when I look at comparable services like CodeBuild, a 2-vCPU instance costs about $0.0034/minute. GitHub is charging only 41% less while the customer is providing the hardware, power, cooling, and network. That doesn't feel proportional, especially when GitHub isn't scaling compute on their end.
A Better Solution: A Self-Hosted Runner Plan
Rather than per-minute billing, what if you introduced a dedicated self-hosted runner plan? Something simple like a $2/month tier that includes unlimited self-hosted build minutes would be ideal. Remember, you're only managing the control plane here, not the actual compute, so $2 feels fair for orchestration and workflow management.
This would appeal to a huge market you're currently underserving: hobbyists, home labs, and small teams who need security isolation but can't justify per-minute costs.
You could sweeten that plan with a few extras to make it feel complete without cannibalising the Pro tier. Things like maybe a higher number of concurrent runners, basic workflow insights, or priority support. Nothing fancy, just enough icing on the cake to make it feel like a solid upgrade.
Even better, offer an annual billing option at a discount, something like $20/year instead of $2/month. People love predictable, annual pricing, and you'd lock in recurring revenue. That works for both sides.
The beauty of this approach is that you convert free users into paying customers reliably. A $2/month tier has a low barrier to entry, especially for people running hobby labs or small infrastructure. You're not asking them to commit to enterprise tiers they don't need.
The Real Issue
I understand GitHub needs to fund operations, but the current model punishes teams for doing the right thing, keeping sensitive infrastructure secure, and for optimising their workflows being efficient. A pricing structure that acknowledges the real difference between fully-managed hosted runners and self-hosted runners would be fairer and probably more profitable for GitHub in the long run.
A self-hosted runner plan solves this elegantly. You get sustainable revenue from a segment that currently pays nothing. Teams get predictable costs and the security they actually need.