Koyeb

What Koyeb actually is

Koyeb is a serverless application platform: push a Git repository or a Docker image and it runs your code in microVMs on top of high-performance bare metal, with autoscaling and billing metered by the second. It is the European entry in the post-Heroku field — the comparisons that come up most are Render and Railway, and the shape is similar. You pick an instance type, a region and a start command; the platform handles the build, the deployment, TLS, load balancing and logs. Anything with a Dockerfile or a supported buildpack can run, so the language question answers itself.

The company behind it is Koyeb SAS, registered in Boulogne-Billancourt near Paris and founded in 2020 by three former Scaleway engineers — Yann Léger, Edouard Bonlieu and Bastien Chatelard. It raised about $8.6 million in total: a $1.6 million pre-seed in 2020 and a $7 million seed in 2023 led by the Paris VC firm Serena. That independence ended on 17 February 2026, when Mistral AI announced its first-ever acquisition: Koyeb. The small team (13 people at the time of the deal) and the three co-founders joined Mistral's engineering group, and Koyeb's platform is expected to become a core component of Mistral Compute, Mistral's AI cloud. Both companies have said the platform keeps operating for existing and new customers in the meantime. We put this up front because it is now the biggest variable in choosing Koyeb: the roadmap serves Mistral's cloud ambitions first, and yours second.

Under the hood, the platform has grown toward AI workloads. There are CPU instance types, a wide GPU lineup (Nvidia L4 up to H200 and B200), Serverless Postgres in Washington DC, Frankfurt and Singapore, and Koyeb Sandboxes — isolated environments for running AI agents safely. For ordinary web services the story stays simple: Docker or Git in, an HTTPS endpoint out, custom domains, health checks, metrics and real-time logs included. Deployment options include single-region or multi-region services with a per-region instance type, which is more control than most PaaS offerings hand you.

Pricing: what you really pay

There are two layers. Compute is metered per second, and the formula is published: number of instances multiplied by instance time in seconds multiplied by the instance price. On top of that sit team plans. Starter is pay-as-you-go with no monthly fee. Pro costs $29 a month, includes $10 of compute usage and business-hours support. Scale costs $299 a month, includes $100 of compute usage, adds direct Slack support, and is the first tier with an uptime SLA. A Business plan with 24/7 support is sold by conversation, and Enterprise is custom — third-party reports put it from around $1,000 a month with private dedicated locations, SSO/RBAC and a 99.99% SLA, though we could not confirm those figures on Koyeb's own pages.

Instance pricing is fully public, and this is Koyeb's strongest card. Standard instances: nano (0.25 vCPU, 256MB RAM, 2.5GB SSD) at $0.0036 an hour, which is $2.68 a month always-on; small (1 vCPU, 1GB, 10GB SSD) at $0.0144 an hour, $10.71 a month; medium (2 vCPU, 2GB, 20GB SSD) at $21.43 a month; and up to 5xlarge (40 vCPU, 128GB, 400GB SSD) at $1,371 a month. Eco instances — available only in Washington DC, Frankfurt and Singapore — start at eco-nano for $1.61 a month (0.1 vCPU, 256MB, 2GB SSD). GPU instances run from an RTX 4000 SFF Ada at $0.50 an hour ($375 a month) and an L4 at $0.70 an hour, through A100 at $1.60 an hour, H100 at $2.50, H200 at $3.00 and B200 at $5.50, up to multi-GPU shapes like 8x H200 at $24 an hour ($17,856 a month).

The free tier is real but small: one free web service per organisation with 0.1 vCPU, 512MB of RAM and 2GB of SSD in Frankfurt or Washington DC, plus one free Postgres database limited to 5 hours of active time and 1GB of storage. Koyeb's own docs say free instances are for testing and hobby projects, not production. The $5.50 monthly credit from the 2022-2023 public preview was discontinued in February 2024 and replaced by the free instance type.

Edge costs are worth reading before you deploy. Outbound bandwidth includes 100GB a month free; the documented rate beyond that is $0.04 per GB, though Koyeb states it is not being charged yet. Custom domains are free up to 100 on Pro and 500 on Scale, with extras planned at $0.20 a month. A credit card is required even to start — Koyeb places a $29 pre-authorization and then charges the prorated plan fee — and there is no spending-limit feature yet; budget caps are still on the public roadmap. Per-second billing plus no hard cap is a combination that deserves a calendar reminder in week one.

Performance & infrastructure

Instances are microVMs — a thin, isolated wrapper around your container — running on high-performance bare metal in seven locations across five countries: Paris, Frankfurt, N. Virginia, San Francisco, Washington DC, Tokyo and Singapore. Placement is tiered: free instances only run in Frankfurt and Washington DC, eco instances in Washington DC, Frankfurt and Singapore, and GPU instances in specific regions. Serverless Postgres covers Washington DC, Frankfurt and Singapore. Each region can run its own instance type for multi-region services.

The cost behaviour is the interesting part. Billing starts when an instance reaches Starting status and stops when the service is paused or stopped, so a side project that scales to zero and back costs almost nothing — as opposed to a per-month PaaS where an idle app still burns $7 to $20. Autoscaling is built in rather than sold as an add-on.

On reliability we keep to what is documented: the SLA defines an Availability Rate per subscription with service credits as the sole remedy, and third parties report 99.9% on the Scale plan and 99.99% on Enterprise. Nothing binds Koyeb below Scale. We have not benchmarked Koyeb ourselves — our benchmarks page covers raw cloud VMs from Hetzner, DigitalOcean, UpCloud, Vultr and Linode — so treat vendor speed claims as claims.

Where it falls short

The ownership change dominates. Mistral AI acquired Koyeb in February 2026, and while both companies say the platform continues, Koyeb's technology is being folded into Mistral Compute and its engineering attention will follow that. Small acquisitions can keep their product promises and still change direction within a year. Third-party reporting suggests new sign-ups are being steered toward paid plans; we could not confirm that from Koyeb's own pages, but we would read the current terms before committing production workloads.

Support is the second gap. Pro gets business-hours help, Scale adds Slack, and 24/7 requires the Business plan. For a platform that sells itself on running production apps, an SLA that starts at $299 a month is a real threshold.

Third, immaturity in the edges: no spending limits yet, outbound bandwidth and extra domains priced but not charged (a bill that can be re-priced later), and modest region coverage compared with the giants. The docs are good but the community is small next to Render or Railway, which matters when you hit a strange build error at midnight.

Finally, enterprise features — private locations, SSO/RBAC, custom agreements — exist only on the custom tier, and its pricing is not public.

Who should pick this

We would pick Koyeb when per-second billing and instant deploy matter more than vendor size: Docker-first teams, side projects that should cost cents when idle, APIs with spiky traffic, and AI teams that want A100 to H200 GPUs with autoscaling instead of a reserved cluster. The French registration is a genuine differentiator for EU data-sensitivity workloads — the company is not subject to the US CLOUD Act — and the eco line (from $1.61 a month) is among the cheapest ways to run an always-on service in a European region.

We would skip it where an uptime guarantee must be contractual on modest spend, where 24/7 support is non-negotiable, or where you want a five-year vendor horizon you can bank on. In those cases a larger provider — or a plain VM on a Hetzner-class host, per our benchmarks — is the safer base.

Key takeaways

• Serverless PaaS with per-second billing: nano instances from $2.68 a month always-on, eco from $1.61 a month, GPU from $0.50 an hour.
• Mistral AI acquired Koyeb in February 2026; the platform continues but now serves Mistral Compute's roadmap — weigh that before betting production on it.
• Free tier exists (one 0.1 vCPU web service in Frankfurt or Washington DC plus a small Postgres), clearly aimed at testing rather than production.
• SLA starts at the Scale plan ($299 a month, reportedly 99.9%); Pro support is business-hours only.
• Seven regions across five countries, French company, EU jurisdiction — decent for a platform this size.
• No spending caps yet and unbilled-but-priced extras (bandwidth, domains) mean the invoice can move; set your own limits.

Sources

• https://www.koyeb.com/docs/faqs/pricing
• https://www.koyeb.com/docs/reference/instances
• https://www.koyeb.com/docs/legal/terms
• https://www.koyeb.com/docs/legal/sla
• https://techcrunch.com/2026/02/17/mistral-ai-buys-koyeb-in-first-acquisition-to-back-its-cloud-ambitions
• https://cloudspress.com/companies/koyeb
• https://deploywhere.dev/platforms/koyeb
• https://hostprodigy.com/host/koyeb