Review

Koyeb Review 2026: Pricing, Pros, Cons and Alternatives

What Koyeb's per-second CPU and GPU instances cost on top of the Pro plan, what the Mistral AI deal changed for new sign-ups, and how Render, Railway and Northflank price the same workload.

Product links may be affiliate links. How we rate 4 alternatives covered
TL;DR

Koyeb is worth it in 2026 for small production APIs and GPU inference that need per-second billing, as long as you accept a monthly plan fee as the floor. Since its February 2026 agreement to join Mistral AI, new users must subscribe to Pro at $29/mo, which includes $10 of compute. A 1 vCPU, 2 GB Eco instance then costs $10.71/mo and an H100 $2.50/hr.

Two small instances on Pro come to about $40 a month by our arithmetic, below Render Pro for the same sizes. Look elsewhere if you want a $0 entry plan (Render, Northflank) or usage-only billing (Railway).

Key factsBest for: Teams that want per-second billing on CPU and GPU…
  • Updated: November 20, 2026
  • Best for: Teams that want per-second billing on CPU and GPU instances, with scale-to-zero for inference.
  • Price as of November 20, 2026: Free plan · Pro $29/mo · Scale $299/mo · custom-priced plans too
  • A serverless CPU and GPU platform with per-second pricing and a paid entry plan for new users.
  • Entry plan: New sign-ups start on Pro, which includes $10 of compute each month.
  • Billing: Instances are billed by the second, with Eco sizes cheaper in three regions.
  • Free resources: One free web instance and one free Postgres database per organization.
  • Ownership: Koyeb agreed in February 2026 to join Mistral AI and keeps operating its platform.

Pros

  • Per-second billing on every instance, with scale-to-zero on paid plans.
  • H100 and A100 hourly rates below Northflank's listed GPU prices.
  • Eco instances give 1 vCPU and 2 GB for $10.71/mo in three regions.

Cons

  • New users must buy the Pro plan; the free Starter plan is closed to new sign-ups.
  • No spending limits yet, so autoscaling needs careful max settings.
  • Platform is moving into Mistral Compute, with the deal subject to closing conditions.
Entry planNew sign-ups start on Pro, which includes $10 of compute each month.
BillingInstances are billed by the second, with Eco sizes cheaper in three regions.
Free resourcesOne free web instance and one free Postgres database per organization.
OwnershipKoyeb agreed in February 2026 to join Mistral AI and keeps operating its platform.

Koyeb is a serverless platform that runs your container or Git repo as microVMs on bare-metal servers, scales them with traffic and bills the seconds they run. Its free Starter plan used to be the way indie developers tried it. That pitch is gone for new accounts: the February 2026 announcement that Koyeb is joining Mistral AI also said new users can only sign up for Pro and above.

This review answers what Koyeb costs after that change, using its pricing page, instance price table and pricing FAQ, all checked on November 5, 2026. We did not deploy a test app; the cost scenarios are arithmetic on list prices.

Toolradar data: of the 54 serverless tools Toolradar, the software directory we run, compares Koyeb with, 85% have a free plan (November 2026). A paid entry ticket for new accounts puts Koyeb in the paid minority, so the per-second rates have to earn that floor back. The same Toolradar page gives Koyeb a 4.6 out of 5 user rating across 21 ratings collected on the web, joint #1 among the 21 serverless tools with 20 or more reviews.

What is Koyeb?

Koyeb is a serverless compute platform for web services, workers, Postgres and GPU workloads. You push a Docker image or connect a GitHub repo, pick an instance size and regions, and Koyeb builds, deploys and load-balances it. Autoscaling, scale-to-zero and per-second billing are the headline features on the pricing page, and the company positions itself for AI work: inference on GPUs, sandboxes for agents and MCP servers.

The plan sets support, seats and limits; compute is billed on top. Pro bundles a slice of compute with 10 users, 100 services, NVMe volumes and snapshots, five concurrent builds, and email and chat support. Scale costs $299/mo with $100 of compute included, 50 users, 1,000 services, AWS regions, a Slack channel and a 99.9% uptime SLA. Enterprise starts at $1,000/mo and adds SSO, RBAC, audit trails, private locations and a 99.99% SLA.

The Mistral AI deal is the context every buyer should know. Koyeb announced on February 17, 2026 a definitive agreement to join Mistral AI, with the whole team moving to Mistral Compute and completion subject to closing conditions. The platform keeps running and Koyeb still bills customers directly, and existing organizations keep their current plan. The free Starter plan is being removed for new users, data is not being moved to Mistral for now, and Koyeb says it will focus on inference, sandboxes and serverless MCP hosting.

How Koyeb works

Compute is priced per instance size, per second. The formula on Koyeb's FAQ is number of instances times seconds running times the size's price. Standard instances run from nano (0.25 vCPU, 256 MB) at $2.68/mo to 5xlarge (40 vCPU, 128 GB). Eco instances give more RAM per dollar: eco-medium has 1 vCPU and 2 GB for the same monthly price as the Standard small with 1 GB, but Eco runs only in Washington, D.C., Frankfurt and Singapore.

GPUs follow the same per-second model, which is the strongest argument for Koyeb in 2026. An L4 costs $0.70/hr, an A100 $1.60/hr and a B200 $5.50/hr, with multi-GPU sizes up to 8x H200. Scale-to-zero means an inference endpoint that sits idle overnight stops billing, which a reserved GPU server cannot match. Long-term reservations can cut on-demand prices by up to 50%, by quote.

Billing has edges worth knowing before the first invoice. Sign-up places a pre-authorization hold equal to the Pro fee on your card and charges a prorated Pro fee for the rest of the month. Outbound bandwidth includes 100 GB a month; Koyeb lists $0.04/GB beyond that but says it is not charging it yet and will give notice. There are no spending limits yet, so an autoscaling service with a high max can surprise you. If a payment fails, new resources can be blocked at 7 days, instances pause at 14 days and deletion starts at 21 days.

The free resources survive inside paid accounts. Each organization gets one Free Instance (0.1 vCPU, 512 MB, 2 GB SSD) in Frankfurt or Washington, D.C., which sleeps after an hour without traffic and cannot run workers or volumes, plus one free Postgres database with 5 hours of active time and 1 GB of storage. They suit a staging preview, not production.

Koyeb key features

Per-second instance billingEssential
CPU and GPU instances bill by the second while running, from $1.61/mo for eco-nano to multi-GPU H200 nodes.
Serverless GPUs with scale-to-zeroEssential
L4, A100, H100, H200 and B200 instances priced per hour on the pricing page, with autoscaling and scale-to-zero for inference endpoints.
Eco instance family
Eco sizes double the RAM of the matching Standard price point, limited to Washington, D.C., Frankfurt and Singapore.
Serverless Postgres
Managed Postgres billed by the second, with scale-to-zero, storage at $0.50 per GB a month and one free 5-hour database per organization.
Git or container deploys
Deploy from a GitHub repository or a prebuilt container image; Pro allows five concurrent builds and NVMe volumes with snapshots.
Sandboxes and MCP hosting
Koyeb says it will focus on sandboxes for agents and serverless MCP servers as it becomes part of Mistral Compute.

Koyeb pricing

All prices below were checked on November 5, 2026 on Koyeb's pricing page, its instances reference and its pricing FAQ, with the plan change for new users from Koyeb's Mistral AI announcement. Taxes are added on the invoice. Monthly figures come from Koyeb's own price-per-month column.

Three monthly scenarios, our arithmetic on those list prices. A side project on one Standard small instance pays $29.71: the Pro fee, plus the $0.71 of compute above what the plan includes. A small production API on two eco-medium instances (1 vCPU, 2 GB each) pays about $40.42. A team on Scale running six Standard medium instances pays about $327.58, because the $100 of included compute covers most of the roughly $128.58 in instance time.

Against rivals for the same 1 vCPU, 2 GB size running all month: Render charges $25/mo per instance plus the same again for its Pro workspace, so two instances cost $75. Northflank charges $24/mo per nf-compute-100-2 instance with no plan fee, so $48 for two. Railway bills $0.00000772 per vCPU-second and $0.00000386 per GB-second, about $40 a month per fully busy instance on a 30-day month, but less when the service is idle because it bills actual use. On GPUs, Koyeb's H100 rate undercuts Northflank's $2.74/hr.

PlanPriceBest for
Koyeb Pro$29/mo + compute$10 of compute included, 10 users, 100 services, volumes and snapshots, email and chat support
Koyeb Scale$299/mo + compute$100 of compute included, 50 users, 1,000 services, AWS regions, Slack support, 99.9% SLA
Koyeb EnterpriseFrom $1,000/mo + computePrivate locations, unlimited users, SSO, RBAC, audit trail, 99.99% SLA, 24/7 support
Standard small instance$10.71/mo ($0.0144/hr)1 vCPU, 1 GB RAM, 10 GB SSD, available in all standard regions
Eco medium instance$10.71/mo ($0.0144/hr)1 vCPU, 2 GB RAM, 16 GB SSD, Washington D.C., Frankfurt and Singapore only
Standard medium instance$21.43/mo ($0.0288/hr)2 vCPU, 2 GB RAM, 20 GB SSD
H100 GPU instance$2.50/hr80 GB VRAM, 15 vCPU, 180 GB RAM, 320 GB disk
Render Pro + 1 CPU, 2 GB service$25/mo workspace + $25/mo per instanceUnlimited seats, 25 GB bandwidth, autoscaling, full-stack previews
Northflank nf-compute-100-2$24/mo ($0.03333/hr)1 vCPU, 2 GB, pay as you go with no seat or plan fee
Railway Hobby$5/mo minimum usageUsage up to the minimum is included, then per-second CPU and memory billing
Heroku Basic dyno$7/mo0.5 GB RAM, always on, free SSL, 10 process types

Koyeb pros and cons

What we like

  • Per-second billing on every instance, with scale-to-zero on paid plans.
  • H100 and A100 hourly rates below Northflank's listed GPU prices.
  • Eco instances give 1 vCPU and 2 GB for $10.71/mo in three regions.

What could be better

  • New users must buy the Pro plan; the free Starter plan is closed to new sign-ups.
  • No spending limits yet, so autoscaling needs careful max settings.
  • Platform is moving into Mistral Compute, with the deal subject to closing conditions.

Who Koyeb is for

Koyeb fits teams deploying AI inference or small APIs that want per-second billing on both CPU and GPU from one provider. If an endpoint sits idle for hours, scale-to-zero and per-second GPU rates beat a reserved server, and the H100 rate is below Northflank's. Startups can also apply for up to $30k in credits, which changes the maths for a first year.

For a plain web app the case is narrower. The Pro floor pays for itself only once you run more compute than the plan includes, and below that a $0 entry plan wins. If you only need one small always-on service, Render on Hobby or Railway on usage billing will usually cost less; our Render pricing guide and Railway pricing guide run those numbers. Get the Devshot brief if you want the next price change on these hosts in your inbox.

Skip Koyeb if you need a hard spending cap today, if your users sit far from its regions, or if the Mistral transition worries your procurement team. Fly.io is another per-second machine host, Modal and RunPod are GPU-first alternatives, and our serverless platforms ranking and Heroku alternatives list cover the wider field.

Best Koyeb alternatives

If Koyeb is not the right fit, these are the closest options.

ToolBest forStarts at
KoyebTeams that want per-second billing on CPU and GPU instances, with scale-to-zero for inference.Free plan · Pro $29/mo · Scale $299/mo · custom-priced plans tooVisit →
RenderWeb apps with a worker and Postgres that want fixed monthly instance prices and a free workspace plan.Free plan · Pro $25/mo · Scale $499/mo · custom-priced plans tooVisit →
RailwayDevelopers whose services sit idle often and who want to pay only for CPU and RAM used.Free plan · Hobby $5/mo · Pro $20/mo · custom-priced plans tooVisit →
NorthflankTeams that want no plan fee, no seat fees and the option to run in their own cloud.Free Sandbox (up to $20/mo of resources)Visit →
HerokuTeams that want the classic Git-push dyno model and Heroku's add-on marketplace.Eco $5/mo · Basic $7/mo · Standard-1x $25/mo · custom-priced plans tooVisit →
Lowest published monthly price by toolBar chart of the lowest monthly price each vendor publishes; the same figures are in the comparison table above.Lowest published monthly priceKoyebFreeRenderFreeRailwayFreeNorthflank$20
Lowest monthly figure each vendor publishes, checked Nov 2026. A tilde marks a figure the vendor states approximately. Per-seat and usage charges can sit on top of it. 1 of 5 does not publish a comparable monthly price and is left out rather than estimated.
Render
Fixed-size instances on a free or paid workspace plan, simple to budget but billed while idle.
Visit →
Railway
Usage-only billing on CPU seconds and memory, with the plan fee counted as included credit.
Visit →
Northflank
Pay-as-you-go containers and GPUs with no plan fee, plus a bring-your-own-cloud option.
Visit →
Heroku
The original platform-as-a-service, priced per dyno with fixed monthly rates and small RAM per tier.
Visit →

The bottom line

Koyeb is a good buy for GPU inference and small APIs that benefit from per-second billing, and a weaker one for hobby apps since the Mistral AI deal. The Pro floor, with its included compute, is fair once you run two or more instances, and the Eco sizes and GPU rates are among the lowest in this group. Confirm your regions, set autoscaling maximums with care, and watch for news as Koyeb moves into Mistral Compute.

Choose Render for fixed prices and a free workspace, Railway for usage-only billing on idle-heavy services, Northflank for no plan fee and bring-your-own-cloud, and Heroku for the classic dyno model. Toolradar profiles Render, Railway and Northflank, and our AI agent sandboxes ranking covers Koyeb's newest use case.

Methodology: Koyeb's plans, instance prices and billing rules were checked on November 5, 2026 on its pricing page, instances reference and pricing FAQ, and rivals' prices on the Render, Railway, Northflank and Heroku pricing pages the same day. Monthly totals are our arithmetic on list prices before tax. Devshot does not deploy test apps for reviews and gives no star rating.

Subscribe free to get the next hosting price change in the Devshot brief.

Cite this: Devshot, "Koyeb Review 2026: Pricing, Pros, Cons and Alternatives", November 2026.

Frequently asked questions

How much does Koyeb cost in 2026?

New users pay $29 a month for Pro, which includes $10 of compute, then per-second compute on top. Scale costs $299 a month with $100 of compute included, and Enterprise starts at $1,000 a month. Instances run from $1.61 a month (eco-nano) and GPUs from $0.50 an hour. Prices were checked on Koyeb's pricing page on November 5, 2026.

Does Koyeb still have a free tier?

Partly. Koyeb's February 2026 announcement says new users can only sign up for Pro or above, and the free Starter plan is being removed for them; existing Starter organizations keep it. Every organization still gets one Free Instance with 512 MB of RAM in Frankfurt or Washington, D.C., and one free Postgres database with 5 hours of active time.

What does the Mistral AI deal mean for Koyeb users?

Koyeb agreed on February 17, 2026 to join Mistral AI, with completion subject to closing conditions. Koyeb says the platform keeps operating, it still bills customers directly, existing plans do not change and customer data is not being moved to Mistral AI for now. Over time Koyeb is to become a core part of Mistral Compute.

Is Koyeb cheaper than Render?

For the same 1 vCPU, 2 GB size, yes once you run two or more instances: two Koyeb eco-medium instances on Pro cost about $40 a month, against $75 for two Render instances on its Pro workspace, by our arithmetic. Render wins for one small service on its $0 Hobby workspace, and Koyeb's Eco sizes exist in three regions only.

How much does an H100 cost on Koyeb?

An H100 instance with 80 GB of VRAM, 15 vCPU and 180 GB of RAM costs $2.50 an hour on Koyeb, billed by the second. Two, four and eight H100 configurations are listed too, and long-term reservations can take up to 50% off the on-demand price by quote.

What are the best Koyeb alternatives?

Render for fixed monthly instance prices and a free workspace, Railway for usage-only billing on services that sit idle, Northflank for pay-as-you-go containers with no plan fee, and Heroku for fixed-price dynos. For GPU-first work, Modal and RunPod are the closer comparisons.

Sources

Prices and plan details come from each vendor's own pricing page, re-checked by the Toolradar pricing tracker.

Related guides

Some offers on this page may be paid placements or contain affiliate links.

Get the Devshot brief

Free daily newsletter, read in 5 minutes.

Subscribe free