Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- KV Instant uses Quicksilver and the Workers KV API but requires private-beta access. Explicitly specify mode: instant when creating a namespace.
- The vendor reports reads below 2 milliseconds at p99 and 99% of writes replicated to the edge in about 250 milliseconds. This article has not independently measured these figures and does not infer strong consistency or guaranteed application latency.
- Each namespace holds at most 1 MB total and 10000 key-value pairs, with keys up to 300 bytes. The write rate is one per second per namespace.
- Announced pricing is $0.20 per million requested keys for reads, $0.10 per put/delete/list operation, and $100 per MB per month for storage. Do not misread the latter two as per million operations or per GB.
- Metadata is unsupported, and list has no pagination. Inventory data size, writes, and inspection frequency before trying a small configuration in isolation. Ordinary KV quotas and pricing do not apply directly.
Private-beta configuration storage, not a universal upgrade to ordinary KV
Cloudflare announced Workers KV Instant private beta on October 1, 2026. It uses Quicksilver for global configuration distribution, targeting feature flags, routing settings, and application configuration that change infrequently but receive many reads. The announcement offers a private-beta application path without promising immediate approval for every account or providing a GA date.
The announcement reports p99 reads below 2 milliseconds and 99% of writes replicated to the edge in about 250 milliseconds. These are Cloudflare's service measurements, not this article's independent benchmark, and not guarantees of application end-to-end latency, strong consistency, or completion time for every write. Account access, actual bills, and real cross-region results all need separate verification.
Limits are per namespace: check update hotspots first
The announced Instant limits are a total of 1 MB and 10000 key-value pairs per namespace. Each key can be at most 300 bytes, and values cannot push the whole namespace beyond capacity. The write limit is one write per second per namespace, not per key. Alternating writes among multiple configuration keys still shares the same namespace-wide constraint.
By comparison, current ordinary KV limits of one write per second to the same key, keys up to 512 bytes, and values up to 25 MiB belong to classic KV. Do not apply ordinary KV's large-value limits to Instant or treat MB and MiB as identical units. Large numbers of orders, frequently changing counters, or large objects do not fit the target pattern in this announcement.
Calculate three costs separately, especially Class A
The Instant announcement prices Class B (get) at $0.20 per million requested keys, Class A (put, delete, list) at $0.10 per operation, and storage at $100 per MB per month. These are announced US-dollar rates, not an account bill inspected by this article. After approval, check current official guidance for actual private-beta billing arrangements and included charges. This article found no separately verifiable public Instant-specific pricing documentation and does not substitute the classic pricing page.
A multi-key get is metered by requested key count: one get for three keys and three single-key get calls both count as three Class B operations. Each key written or deleted counts as one Class A operation. One list counts as one Class A operation regardless of the number of keys returned. At the announced rate, 1000 Class A operations arithmetically cost $100. This is not a complete monthly bill and excludes plans, other services, taxes, and discounts. Frequent list-based inspections can also incur charges.
Cheaper reads do not make the entire solution cheaper
The classic KV pricing page lists reads at $0.50 per million, writes/deletes/lists at $5 per million, and storage at $0.50 per GB-month, with included allocation by plan. These units differ from Instant's per-operation Class A and per-MB-month storage. Comparing only $0.20 with $0.50 cannot establish total savings from migration.
First list expected requested keys, put/delete key counts, list calls, storage size, and update peaks, then meter each item. The Instant announcement does not grant classic's daily free quota or paid included allocation, and this article does not transfer them. If frequent writes or large storage dominate the workload, the announcement itself suggests continuing to consider ordinary KV.
Similar API, different metadata and list behavior
Specify mode: instant when creating an Instant namespace. An existing classic namespace does not become Instant merely by upgrading an SDK. The announcement says most binding operations are compatible, but metadata is unsupported: getWithMetadata returns null metadata, and put cannot accept metadata. Applications using metadata for versions, types, or expiration information should check their logic first.
Instant list returns all matching keys on one page without pagination. Migrated code must not keep waiting for classic's next-page cursor or interpret null metadata after migration as a deletion signal. Similar API shapes do not make behavior and costs interchangeable. Use a separate namespace and nonsensitive configuration for the first test.
After approval, validate a small configuration and retain a fallback
Confirm access conditions through the announcement's private-beta entry point, then follow the official namespace-creation method after approval. Explicitly set instant mode during testing and write only a few test keys. Verify single- and multi-key get, absent metadata, unpaginated list, and whether updates across multiple keys trigger namespace-wide limits. Record version markers, regions, and write/read times to distinguish application caching, network latency, and service replication. This article has not created namespaces or executed these requests.
Before migration, retain original classic data and read paths, and identify who switches bindings and rolls back. For write throttling, first reduce updates across the whole namespace. Stop switching if access is unapproved, costs are unclear, or existing code depends on metadata or pagination. Use the proxy connection troubleshooting checklist only with independent network evidence such as connection timeouts, TLS, or proxy authentication errors. Network tools cannot change beta eligibility, capacity, operation prices, or write limits.
Sources
Frequently Asked Questions
Is KV Instant GA or open to all accounts?
No. The October 1, 2026 announcement explicitly calls it private beta and provides an application path. This article has not verified a specific account's approval, a GA date, or actual bills.
Is Instant's $0.10 price per million operations?
No. Announced Class A pricing for put/delete/list is $0.10 per operation. Only Class B reads are metered at $0.20 per million requested keys.
Is $100 per MB-month a typo for GB?
The official announcement read for this article explicitly says $100 per MB per month. This article retains that unit instead of changing it to GB. After approval, still check current official prices and account billing arrangements.
Does one write per second apply to a key or a namespace?
For Instant, it applies to the entire namespace. The corresponding classic KV limit is one write per second to the same key. Multiple keys share an Instant namespace's write constraint.
Can ordinary KV metadata, pagination, and free quotas migrate directly?
No. Instant does not support metadata, and list has no pagination. Classic quotas and storage units cannot replace Instant's announced rules.
Does 250-millisecond replication imply strong consistency or that every request is that fast?
No. Cloudflare reports that 99% of writes replicate in about 250 milliseconds. This article has not independently tested it and does not turn that percentile into a consistency or end-to-end latency guarantee.