Cloudflare launches Workers KV Instant for sub-millisecond global reads

Cloudflare has introduced Workers KV Instant, a new operational mode for its key-value store that leverages the internal Quicksilver system to achieve sub-millisecond read latencies. Announced on October 1, 2026, this update allows developers to access globally replicated data with significantly reduced delay compared to the classic Workers KV model. The service is currently available in private beta for selected customers.

What happened

Workers KV has been a staple of Cloudflare’s developer platform since its launch in 2018, serving as a reliable store for static assets and user configurations that are read frequently but updated occasionally. However, the classic mode involves eventual consistency, meaning changes can take seconds to propagate across the global network. For years, developers have requested access to Quicksilver, Cloudflare’s internal high-speed key-value store that powers nearly every request to its network, but it remained exclusive to internal use until now.

With the release of Workers KV Instant, Cloudflare is exposing the underlying Quicksilver v2 infrastructure to external users through the familiar Workers KV API. This new mode is designed specifically for infrequently updated application configuration data, such as feature flags or homepage settings, where immediate global availability is critical. Unlike the classic mode, which may require waiting for time-to-live (TTL) expiration to see updates, KV Instant pushes changes to the edge instantly, eliminating cold read penalties and ensuring that data is available globally within milliseconds of being written.

How it works

Workers KV Instant operates by routing data through Quicksilver v2, a distributed key-value store engineered for planet-scale replication and low-latency access. When a write operation occurs, the data is propagated to over 300 edge locations worldwide. The system achieves a p99 write replication time of approximately 256 milliseconds, meaning 99% of writes are visible globally in under a quarter of a second. Read operations are even faster, with p99 latencies under two milliseconds and p95 latencies measured in microseconds. This performance profile makes it suitable for reading data in the hot path of applications without introducing noticeable delay.

The API remains largely consistent with the existing Workers KV interface, supporting standard operations like get(), put(), list(), and delete(). Developers can integrate it into their workflows with minimal code changes, such as checking a feature flag on every request to determine which version of a homepage to serve. However, there are specific technical constraints: metadata support is removed, so getWithMetadata returns null, and list operations return all matching keys without pagination. Additionally, namespaces must be explicitly created with the mode: "instant" attribute to activate these performance characteristics.

Key details

• Read Performance: p99 read latency is under 1.62 ms, which is over 100 times faster than the classic mode’s 287 ms p99.
• Write Replication: 99% of writes replicate to all edge locations in approximately 256 ms, compared to up to 4.38 seconds in classic mode.
• Storage Limits: Namespaces are limited to 1 MB total size, with individual keys up to 300 bytes and a maximum of 10,000 key-value pairs per namespace.
• Write Throttling: Writes are restricted to one operation per namespace per second to maintain consistency and order.
• Pricing Structure: Reads cost $0.20 per million (60% less than classic), while storage is priced at $100 per MB per month, and Class A operations cost $0.10 each.
• Availability: The service is launching in private beta, with Cloudflare inviting developers to sign up and provide feedback on their use cases.

Why it matters

For software engineers building global applications, the trade-off between consistency and latency has often required complex caching strategies or third-party services. Workers KV Instant simplifies this by providing strong consistency guarantees with near-instant propagation. This is particularly valuable for time-sensitive operations, such as coordinating a product launch or updating emergency configuration settings, where delays of several seconds can impact user experience or business outcomes. By removing the need to manage TTLs or wait for eventual consistency, developers can rely on the store for real-time decision-making at the edge.

The pricing model also reflects a shift in how cloud storage costs are structured for high-performance workloads. While read operations are significantly cheaper, the high cost of storage and write operations encourages efficient data design. Developers must be mindful of the 1 MB namespace limit and the per-operation costs for writes and lists. This model favors small, frequently accessed datasets over large bulk storage, pushing teams to optimize their configuration data structures and avoid using the service as a general-purpose database for large objects.

What you can do

• Sign up for the Workers KV Instant private beta through the Cloudflare developer portal to test the new capabilities.
• Audit your current Workers KV usage to identify configuration data or feature flags that would benefit from sub-millisecond read latency.
• Refactor namespace creation scripts to include the mode: "instant" attribute when provisioning new stores for high-priority data.
• Review application logic to remove dependencies on metadata fields, as KV Instant does not support getWithMetadata operations.
• Optimize list operations by assuming they will return all keys in a namespace, and adjust memory allocation in your Workers to handle potentially larger response payloads.
• Calculate expected costs based on the new pricing tiers, focusing on the high cost of storage ($100 per MB) and Class A operations to avoid unexpected bills.