ZBSearchCost calculator
Estimate the monthly Cloudflare bill for a ZBSearch Edge deployment - Workers requests and CPU, R2 storage and operations, and Durable Objects, from measured baselines.
ZBSearch Edge runs entirely on Cloudflare Workers, R2, and Durable Objects (a write coordinator per index, a search node per shard), so the monthly bill is a function of three workload dimensions: documents, searches, and writes. Drag the sliders to match your workload:
ZBSearch Edge is unoptimized right now. Real-world costs are higher than they should be, and we're actively working to make it way cheaper.
Requires Workers Paid ($5/mo base, included above). The free plan caps CPU at 10ms per invocation, which search workloads exceed.
Warm search ~2.0ms CPU; cold isolate adds ~25ms for snapshot load.
How the estimate works
The calculator applies Cloudflare's published pricing (verified July 2026) to resource usage measured with the load-testing harness:
| Input | Value | Source |
|---|---|---|
| Index size | ~1.39 KB/doc | 13.87 MB snapshot per 10k docs, measured on R2 |
| Warm search CPU | ~0.4ms per 1k docs | ~1.1ms @10k, ~40ms @89k docs measured |
| Cold-isolate search | +~5µs/doc (snapshot load) | ~0.5s at 100k docs, assumed 2% of searches |
| Writes | batched 100/request, ~30ms CPU | 75 docs/s via API measured on a deployed Worker |
| Rebuilds | every 500 ops/shard, ~62µs CPU/doc | 96k docs rebuilt in ~6s CPU, measured |
| R2 meta reads | shards+1 GETs per search | meta.json is re-read per shard per search (uncached) |
| Search fan-out | 1 ShardSearch DO request per shard per search | Distributed search: each shard query runs in its own DO isolate |
Pricing dimensions: Workers Paid $5/mo (includes 10M requests + 30M CPU-ms, then $0.30/M requests and $0.02/M CPU-ms), R2 $0.015/GB-month with Class A at $4.50/M and Class B at $0.36/M (10 GB + 1M A + 10M B included, egress free), Durable Objects $0.15/M requests (1M included; DO duration and storage are negligible at these snapshot sizes).
Rules of thumb
- Below ~1M searches/month, everything fits in the $5 base plan. R2 storage and Durable Objects stay rounding errors at any scale on this page.
- The Workers Free plan is not viable for search. It caps CPU at 10ms per invocation - measured 97.8% of searches failing with error 1102 on a 10k-doc index. Writes do fit.
- Enable the meta cache in the calculator to see the single biggest cost lever at high QPS: per-shard
meta.jsonreads currently hit R2 on every search, and at 30M searches/month they cost more than compute. - Sharded searches multiply DO requests. A search on an N-shard index makes N
ShardSearchDO requests - at 5 shards and 30M searches/month that's 150M DO requests (~$22). Still cheaper than the R2 reads the meta cache removes. - CPU scales with corpus × QPS. Each shard DO searches its full snapshot, so past ~100k docs per shard the compute line dominates. Beyond that, route queries to relevant shards/tenants instead of scatter-gathering over everything.
Production
Harden ZBSearch Edge for production - authentication, external rebuilds, GitHub Actions, and operational guidance. Cloudflare deploy is in beta.
Benchmarks
Load-test ZBSearch Edge with the k6 harness in benchmarks/http - corpus generation, timed ingest, search/write/mixed/coldstart scenarios, and CI. Cloudflare deploy is in beta.