Heroku

What Heroku actually is

Heroku is the platform that taught the industry what "git push heroku main" could mean. Founded in San Francisco in 2007 by James Lindenbaum, Adam Wiggins and Orion Henry — in development from June of that year, originally for Ruby — it was one of the first true cloud platforms-as-a-service. Salesforce bought it in December 2010 for $212 million and has owned it since. Today it supports Java, Node.js, Scala, Clojure, Python, PHP, Go and Ruby, runs everything on AWS infrastructure, and organizes compute into "dynos": isolated Linux containers you scale by count and size rather than by managing servers.

The value proposition has never changed much: you push code, the platform builds it, runs it, patches the OS, gives you logs, SSL, custom domains and a large add-on marketplace (Heroku Elements), and you never think about a machine. That is the product we recommend to developers who want to ship, and it is why Heroku survived a decade of cheaper competitors.

But this review cannot skip the current chapter. In August 2022 Heroku discontinued its free plans, citing fraud and abuse — the free dyno era ended that November. In April 2022 the platform suffered a serious security intrusion involving stolen OAuth tokens that reached customer GitHub repositories. In March 2024 Heroku announced a replatform onto Kubernetes. And in February 2026 the company said it was moving to an engineering model focused on security, reliability, support and stability "rather than introducing new features," and that Enterprise Account contracts would no longer be offered to new users. In our reading, that is a platform entering maintenance mode: supported, stable, not racing forward. Everything below is priced against that reality.

Pricing: what you really pay

Heroku's pricing page is refreshingly legible: four dyno tiers, flat numbers, no calculator games.

Eco is $5/month flat for a pool of 1,000 uptime hours shared across your apps, with dynos sleeping after 30 minutes of inactivity — the direct replacement for the old free tier, aimed at testing ideas and intermittently used apps. Basic is $7 per dyno per month (0.5GB RAM, always on, up to 10 process types). Standard runs $25–$50 per dyno per month (Standard-1X: 0.5GB at $25; Standard-2X: 1GB at $50), adding horizontal scaling, preboot, threshold alerting and finer metrics. Performance dynos run $250–$1,500 per dyno per month with dedicated compute: Performance-M (2.5GB) at $250, Performance-L (14GB) and Performance-L-RAM (30GB) at $500 each, Performance-XL (62GB) at $750, Performance-2XL (126GB) at $1,500. Private Spaces and Shield Spaces (network isolation and high-compliance environments) price on top. All figures come from Heroku's dyno pages and the devcenter's dyno size table.

There is no free tier, and the honest comparison is stark: $25/month buys 0.5GB of RAM and a 1x CPU share. On a raw VPS, the same money buys several times the resources — our /benchmarks page has the numbers. What the $25 buys is the platform: builds, patching, routing, logs, add-ons, review apps, and no on-call rotation for the OS. If your team's time is expensive, that arithmetic can still work. If you are comparing RAM per euro, it never will.

Add-ons are metered separately and the marketplace is the ecosystem's real depth — databases (Heroku Postgres, Heroku Redis), monitoring, queues, email, and long-tail services provisioned into your app. Note that add-on pricing and region support vary; latency-sensitive add-ons that are unavailable in your app's region will fail to provision.

Performance & infrastructure

Dynos are virtualized Linux containers with defined memory and CPU share: 0.5GB at the entry tiers, up to 126GB on Performance-2XL. Cedar-generation dynos report CPU as shares (1x to 100x compute); the newer Fir generation reports real vCPU counts (for example dyno-1c-0.5gb at $25/month or dyno-8c-32gb at $500/month, dedicated compute throughout). Scaling is the platform's best trick: change the formation and the change is instant; Standard and Performance tiers add preboot (new dynos accept traffic before old ones shut down) and Performance adds autoscaling.

Deployment regions: the Common Runtime offers just two — us and eu — with us as default. Private Spaces add ten AWS-backed regions: Virginia, Oregon, Dublin, Frankfurt, London, Montreal, Mumbai, Singapore, Sydney and Tokyo. Heroku's own docs caution that naming a region does not guarantee data residency for products, services and add-ons, which matters if compliance drives your region choice. One structural note: you cannot move a Common Runtime app between regions without redeploying; there is no migration path.

Reliability is a mixed record. The platform has run production apps at scale for well over a decade and the operational surface (automated OS patching, unified logs, free SSL with automated certificate management) is mature. Against that: the 2022 breach was serious and disclosed in stages, and the 2026 shift to a stability-first engineering model means the pace of platform improvement you are buying into is slower than competitors who are still shipping. No public uptime SLA applies to Common Runtime tiers as far as we can find.

Where it falls short

Price-performance is the obvious wound, and the 2026 product posture makes it worse: you pay premium PaaS prices for a platform whose forward roadmap is explicitly "keep it running." For greenfield projects in 2026, we would look hard at alternatives before defaulting here — the deploy experience is still excellent, but the cost per unit of compute and the feature velocity both point the wrong way.

The free tier is gone, and the Eco replacement sleeps apps and meters hours; hobby projects now cost money or nap. The 2022 security incident is also part of the record: Heroku's remediation and disclosure were criticized, and teams with strict supply-chain requirements should evaluate the current controls rather than assume.

Lock-in is real but gentle: buildpacks and the twelve-factor model make apps portable in principle, yet add-ons, Heroku Postgres patterns, review apps and the CLI workflow tie daily practice to the platform. Data residency guarantees are weaker than the region picker suggests, and Common Runtime is a multi-tenant world — network isolation means paying for Private Spaces.

On migration economics, our honest view is that the exit cost is rarely the blocker — most twelve-factor apps move to a container platform in days — but the opportunity cost is. Teams leave Heroku and rebuild pipelines, logging, scaling policy and add-on substitutes; that is a project measured in weeks, and it is the real reason platform loyalty persists even when the pricing looks indefensible on paper. Budget for the project explicitly rather than assuming the move is free, and re-run the numbers with the migration included.

There is also the question of what Salesforce ownership means in 2026. It guaranteed survival through the years when independent PaaS vendors disappeared, and the February 2026 announcement reads like a product being stewarded rather than grown. That is a defensible strategy for a platform with a large installed base, and a poor one for a buyer seeking a long-term innovation partner. Judge which of those two you are before you sign.

Finally, the Enterprise direction: existing Enterprise customers are supported, but new Enterprise Account contracts are no longer offered as of February 2026. Large buyers should get current terms in writing before planning a multi-year commitment.

Who should pick this

Our pick for small teams and solo developers whose bottleneck is engineering time, not hosting budget — people shipping a product on Node, Python, Ruby or Go who want git push deploys, a huge add-on ecosystem and zero server operations. Existing Heroku customers with stable apps also fit: the platform runs, the workflow is second nature, and migration is a project with its own cost.

Skip it if you are optimizing spend per GB of RAM, if you need EU data residency guarantees, if you want a platform with a fast-moving feature roadmap, or if you are signing a new enterprise agreement — the door is partly closed. For alternatives at every level, compare /category/managed-cloud; for the raw compute those dollars could buy instead, see /benchmarks.

Key takeaways

• The original PaaS: git push deploys since 2007, Salesforce-owned since its $212M acquisition in 2010.
• Dyno pricing: Eco $5/month flat (1,000 shared hours, sleeping dynos), Basic $7, Standard $25–$50, Performance $250–$1,500 per dyno — the full range is on Heroku's dyno pages.
• No free tier since late 2022; the free dyno era is over.
• Regions: us and eu Common Runtime; ten AWS-backed Private Spaces regions from Virginia to Tokyo.
• February 2026: engineering focus shifted to stability, not new features; new Enterprise contracts paused.
• Best fit: time-poor developers who value the workflow and ecosystem over price-performance.

Sources

• https://www.heroku.com/dynos/
• https://devcenter.heroku.com/articles/dyno-sizes
• https://devcenter.heroku.com/articles/regions
• https://en.wikipedia.org/wiki/Heroku