Cloudflare K2 in beta: serverless event streams on R2 at $0.04/GB
Cloudflare has opened the public beta of K2, a durable event log built on R2: limits of 10 GB and 30 MB/s per stream, with future pricing already published.
UptimeMag editorial team · 1 October 2026 · 2 min read

On 1 October 2026, Cloudflare opened the public beta of K2, a serverless event streaming service designed for those who need to decouple event producers from event consumers without managing a Kafka cluster. For anyone running backends on Workers or CDN setups with real-time analytics and fraud detection logic, this is a new addition to Cloudflare's developer platform, alongside Queues and Pipelines.
As described in the company's official blog, it works like an ordered log: it's fully serverless, scales to handle large volumes of data, and supports long-term retention, so that even extended periods of consumer inactivity don't result in data loss. Under the hood, the system writes a partitioned, durable log on top of R2 object storage, rather than relying on local disk as a traditional Kafka broker would.
Anyone wanting to try it needs a Workers Paid subscription. K2 is available from today in public beta for accounts with Workers Paid subscriptions, subject to the limits shown in the table below.
| Item | Value during beta |
|---|---|
| Maximum storage per account | 10 GB across all streams |
| Maximum produce rate per stream | 30 MB/s |
| Billing during beta | None |
| Price for data produced (post-beta) | $0.04/GB |
| Price for data consumed (post-beta) | $0.04/GB |
| Price for data stored (post-beta) | $0.02/GB per month |
K2 usage won't be billed during the beta period. Once billing begins, Cloudflare expects to apply the prices listed in the table above. Anyone needing higher limits can request them by completing the Limit Increase Request Form on the developer site.
Anyone evaluating the service should factor in the cost of fan-out: with multiple independent subscriptions on the same stream, the consumption cost multiplies for each consumer, since the tariff is applied to data read rather than a fixed fee per shard as with some other cloud streaming services.
On the latency front, the decision to build on R2 involves a trade-off the company openly acknowledges: batches are accumulated in memory before being written as a segment, which results in roughly one second of produce latency at the 99th percentile in the initial release.
For those who only need to write JSON events to R2 or Iceberg tables without custom processing, Cloudflare continues to recommend Pipelines; K2 remains the choice for those doing custom processing or writing to destinations other than R2.
Written with the help of artificial intelligence and checked by the editors (EU AI Act, art. 50).