Algolia vs Typesense Cloud vs Meilisearch Cloud: Real Search Cost at 10 Million Records and 100 Million Queries in 2026
The short answer: the three vendors in this comparison don't even agree on what should scale the bill. Algolia bills per search request and per record, so a genuinely high-traffic workload directly multiplies cost; at 10 million records and 100 million queries a month, Algolia's Grow-tier rate card ($0.50 per 1,000 searches, $0.40 per 1,000 records) works out to roughly $54,000/month. Typesense Cloud bills for provisioned RAM and CPU, not search volume at all, so the same 100 million queries cost nothing extra as long as the cluster's throughput capacity can handle the resulting query rate — an ILLUSTRATIVE estimate for a cluster sized to this record count lands around $200–$250/month, a difference of more than 200x for the search-query line specifically. Meilisearch Cloud's flat Cloud tiers start with a 50,000-search included allowance, far short of 100 million; its real cost at this volume is QUOTE-ONLY / UNKNOWN, since no confirmed overage rate for its Cloud tiers at this scale was found in public documentation.
The billing-unit mismatch, up front
Algolia bills two separate usage dimensions — search requests and indexed records — each with its own per-thousand rate, meaning cost scales directly and continuously with both query volume and index size. Typesense Cloud bills infrastructure — dedicated RAM and vCPU, provisioned hourly, regardless of how many searches actually hit the cluster, as long as it has the capacity to serve them. Meilisearch Cloud bills a flat monthly subscription with an included search allowance and (per vendor positioning) comparatively gentler overage pricing than Algolia's, though the exact overage rate at high volume is not confirmed in public sources. These are three structurally different pricing philosophies, not three vendors competing on the same per-unit rate — a "cost per million queries" framing only makes direct sense for Algolia, since the other two don't meter queries as the primary cost driver at all.
What each vendor bills
Algolia (OFFICIAL, algolia.com/pricing, checked July 2026). Build: free, 10,000 searches/month, 1,000,000 records, 10,000 AI Recommendation requests. Grow (usage-based): $0.50 per 1,000 search requests beyond the free allowance, $0.40 per 1,000 records beyond the free allowance. Grow Plus: a steeper $1.75 per 1,000 searches, positioned for higher-volume or higher-feature needs. A critical, easy-to-miss mechanic: "search-as-you-type" autocomplete fires one billable search request per keystroke unless the integration debounces input, which can multiply effective query volume several times over actual user searches. Record-counting rules also have surprises: standard replicas (used for custom sort orders) duplicate the record count and its associated cost, while "virtual replicas" (Algolia's Relevant Sort feature) avoid that duplication.
Typesense Cloud (OFFICIAL, Typesense's pricing documentation and benchmark data). Billed purely by provisioned cluster resources — RAM and vCPU, by the hour — with no per-search or per-record charge of any kind. A documented reference point: a 0.5 GB RAM / 2 vCPU cluster with a 1-hour daily burst allowance costs roughly $28.80/month. Typesense's own benchmark reports indexing 2.2 million records at roughly 900 MB of RAM while sustaining 104 concurrent queries/second at an average 11 ms response time on a 4-vCPU server — giving a rough capacity reference of about 0.409 KB (roughly 0.0004 MB) of RAM per indexed record for a comparable dataset shape, though actual RAM usage varies significantly by document size and field configuration. Scaled to 10 million records of similar shape, that ratio implies roughly 4 GB of RAM — the basis for the illustrative cluster-sizing estimate below. Because there's no per-query charge, a traffic spike never creates a separate line item — the only cost consequence of more traffic is needing a cluster sized with enough throughput headroom to serve it without degrading latency.
Meilisearch Cloud (OFFICIAL, Meilisearch's pricing page, with some conflicting entry-tier figures across sources). Cloud tiers starting around $20–30/month (sources differ on the exact entry price, possibly reflecting different plan names or recent changes), including 50,000 searches/month — five times Algolia's free-tier search allowance. Beyond that allowance, Meilisearch is described by multiple sources as having "gentler overages" than Algolia, but no specific overage rate is published, so it remains UNKNOWN. A separate resource-based plan (reported around $23/month) bills for dedicated CPU and RAM instead of per-query usage, closer in spirit to Typesense's model. Meilisearch's core engine remains MIT-licensed and self-hostable at any scale, a cost-elimination path neither Algolia nor Typesense Cloud offers in comparable open-source form.
Normalizing the workload: 10 million records, 100 million queries/month
All ILLUSTRATIVE for the Typesense RAM sizing, which depends heavily on actual document size and field structure not specified by a single universal figure.
Formula: Algolia = (records ÷ 1,000 × $0.40) + (searches ÷ 1,000 × $0.50); Typesense = cluster cost scaled from a known reference point by estimated RAM requirement, explicitly labeled as an illustrative extrapolation rather than a vendor-confirmed rate at this exact size; Meilisearch = QUOTE-ONLY / UNKNOWN beyond its 50,000-search included allowance, since no confirmed overage rate at this volume is publicly available.
| Platform | Basis | Monthly cost at 10M records / 100M queries |
|---|---|---|
| Algolia, Grow tier | 10M records × $0.40/1,000 + 100M searches × $0.50/1,000 | $54,000 |
| Typesense Cloud | ILLUSTRATIVE estimate: 4 GB RAM needed (8x the 0.5 GB reference cluster), scaled from the $28.80/0.5GB-RAM reference point (8 × $28.80 ≈ $230) | ~$200–$250 (confirm exact figure via Typesense's own calculator, not a vendor-confirmed rate) |
| Meilisearch Cloud | 50,000-search entry allowance is a tiny fraction of 100 million; no confirmed overage rate at this scale | QUOTE-ONLY / UNKNOWN |
The more-than-200x gap between Algolia and the Typesense estimate at this specific volume is the central structural fact of this comparison: at genuinely high query volume, a per-request pricing model and a provisioned-infrastructure model diverge by orders of magnitude, because one scales with usage and the other does not.
Why autocomplete can silently inflate Algolia's real query count
Formula: effective queries = displayed search results × average keystrokes per completed search, when search-as-you-type is implemented without debouncing. A user typing a 10-character query without any debounce delay can fire up to 10 separate billable search requests for what feels, to them, like a single search — meaning Algolia's "100 million queries" in this article's normalization could represent far fewer actual user searches than the number itself suggests, depending entirely on frontend implementation choices outside Algolia's own pricing page.
Where Typesense's model stops being free of query-count risk
The absence of a per-query charge does not mean query volume is irrelevant to cost — it means the cost shows up as a capacity requirement instead of a line item. A cluster sized for 100 million queries a month (roughly 38 queries/second sustained, averaged across the month) needs enough provisioned vCPU and RAM to serve that throughput without degrading latency. Against Typesense's own benchmark reference (104 queries/second sustained on a 4-vCPU server), a 38 QPS average sits comfortably within what a modestly-provisioned cluster can handle, suggesting the RAM requirement (roughly 4 GB, for record storage) is the binding constraint for this illustrative workload, not CPU or query throughput — though a real deployment should still provision headroom above the average for traffic spikes, since sizing too small doesn't generate an overage bill the way Algolia's model would, but it does degrade the actual search experience, which is a real cost in a different form.
Sensitivity
- Query volume and its effect on each billing model differently. Directly and linearly multiplies Algolia's bill; indirectly affects Typesense's bill only insofar as it determines required cluster throughput capacity.
- Debounce implementation on autocomplete, for Algolia specifically. Can multiply or shrink effective billable search volume by several times for the same actual user search behavior.
- Document size and field configuration, for Typesense's RAM sizing. This article's roughly 0.4 KB/record estimate is dataset-specific; a more complex document schema with more or larger fields requires proportionally more RAM per record.
- Whether Meilisearch's flat-tier overage, once confirmed, is genuinely "gentler" than Algolia's at this volume. Not independently confirmed, but the vendor's own positioning suggests it should land somewhere between Algolia's and Typesense's cost profiles.
Budgeting traps
- Pricing Algolia off demo-scale search volume and extrapolating linearly to production traffic without accounting for autocomplete's per-keystroke billing. Real production query counts are often several times higher than raw user-search counts suggest.
- Assuming Typesense Cloud's lack of per-query pricing means traffic growth is free. It still requires provisioning a larger, more expensive cluster to maintain latency at higher sustained query-per-second rates.
- Treating Meilisearch's "gentler overages" claim as a specific, budgetable rate. No confirmed overage figure exists in public sources at meaningful volume.
- Comparing Algolia's per-unit rate directly to Typesense's or Meilisearch's flat/tiered pricing as if all three were the same billing shape. They are not, and a "per million queries" framing only cleanly applies to Algolia.
What to ask before you buy
Measure your actual search-as-you-type keystroke-to-query ratio before budgeting Algolia at any volume, since unoptimized autocomplete can multiply billed search requests well beyond real user search counts. Ask Typesense Cloud directly for a sized cluster quote at your actual record count and document schema, and ask Meilisearch directly for its current overage rate beyond the included search allowance, since neither figure could be confirmed from public documentation at this article's normalized volume.