Turso
Turso (joining Supabase) · databases · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
Row-based pricing with the largest free cap in the table — 500M reads + 10M writes/mo covers our hero workload 500× over at $0 — now “joining Supabase” (02.10.2026).
#2 of 9 in Databases
ToolScore breakdown — why 88/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸The free tier covers our hero workload 500× over — reads are the unit and you get 500M of them
- ▸At the hero workload Turso is one of three providers (with Neon and Firestore) whose free tier genuinely covers the month at $0
- ▸500M reads + 10M writes/mo · 100 DBs · 5GB — every free cap published
- ▸libSQL is a SQLite fork, and embedded replicas put copies of your data in a local file you hold
- ▸Embedded-replica reads come from the local file in microseconds — local-first reads, managed writes
Where it costs you
- ▸“Turso is joining Supabase” (official blog, 02.10.2026) — the product continues, but deal terms are not published
- ▸Developer plan is $5.99/mo — the $4.99 in their own meta description is stale
- ▸Rows READ are what bills: an unindexed SELECT bills every row it scans
- ▸Embedded replicas: sync ships 4KB pages — a 1-byte change is a full frame; never open the file mid-sync (corruption risk); no filesystem (serverless) = no replica
Hidden traps — the stuff the pricing page doesn't say
Turso bills rows READ — the driver of the bill is a missing index, not storage. A query that scans a million rows bills a million reads; against the 500M free cap that is survivable, against a paid month it is the line item you didn’t model.
“Turso is joining Supabase” (official blog, 02.10.2026) — the product continues, but deal terms are not published. For a database choice, that is an open variable to price in, not a reason to walk away.
Sync ships 4KB pages — a 1-byte change is a full frame — and opening the file mid-sync risks corruption. On serverless runtimes there is no filesystem at all, so there is no replica.
Turso’s own meta description still says $4.99; the Developer plan is $5.99/mo (2.5B reads / 25M writes). Budget $5.99.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official siteAffiliate link where available — never affects the score.
- official Turso pricing — official, checked 2026-11-12 verified
- official “Turso is joining Supabase” — official blog, 02.10.2026 verified
- official Turso embedded replicas docs — sync pages, corruption warning verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
Single file, zero server, public domain — 10–50K reads/s on NVMe and two production radars run on it
The BaaS whose exit is pg_dump — Supabase → any Postgres in 1–5 days, zero schema change — $25/mo Pr
Postgres on a Hetzner CX23 (2 vCPU / 4GB / 40GB, 20TB traffic) at €5.99/mo ex. VAT — the box that be
Turso review
In-depth analysis · last updated 2026-11-12
The row-based free-tier king
Turso bills by rows read and written, and the free tier that comes with that model is the largest in this comparison: 500M reads + 10M writes/mo, 100 databases, 5GB of storage. Our hero workload — 1M reads + 100K writes — is 0.2% of the free read cap: the hub’s own line is that the free tier covers the workload 500× over, and Turso is one of the three providers (with Neon and Firestore) whose free tier genuinely covers the month at $0. The unit is also the trap: rows READ are what bills, and an unindexed SELECT bills every row it scans.
The paid step — and the acquisition
Developer is $5.99/mo for 2.5B reads / 25M writes — while the $4.99 in Turso’s own meta description is stale, so budget $5.99. The bigger open question is ownership: “Turso is joining Supabase” (official blog, 02.10.2026) — the product continues, but deal terms are not published. For a database decision — the kind you marry — that is a real, dated uncertainty, and the hub carries it as a warning rather than a disqualifier.
Embedded replicas — the second product on the same account
The hub’s self-hosted table scores Turso embedded replicas separately, at 72: reads come from the local file in microseconds, writes sync to a managed primary with read-your-writes immediately. The mechanics that decide whether that mode is for you: sync ships 4KB pages — a 1-byte change is a full frame; never open the file mid-sync (corruption risk); and where there is no filesystem (serverless), there is no replica. libSQL is a SQLite fork, so the whole stack — local reads, managed writes, export — stays in the SQLite lineage.
How the pillars land
Price/Value 10: $0 at the hero workload with 500× headroom — nothing else in the managed table combines those two. Real limits 10: 500M reads + 10M writes/mo · 100 DBs · 5GB, every cap published. Latency & cold starts 9: embedded-replica reads from the local file in microseconds; no managed-latency numbers are published. Data handling 7.5: the one fact is that embedded replicas keep copies of your data in a local file you control — beyond that, no data-handling facts in this snapshot. Portability & export 9.5: libSQL is a SQLite fork and the local file is the export. Feature set 9: 100 DBs on the free tier, embedded replicas, and the simplest billing unit in the table. Track record 7: the Supabase join is announced, the terms are not. Ease of use 9: “reads are the unit and you get 500M of them” — a mental model a beginner can hold. Pricing transparency 8.5: the unit math is published; their own meta description still says $4.99. Support & docs 8: the embedded-replicas docs are load-bearing — the mid-sync corruption warning comes straight from them.
Verdict — FREE 500× HEADROOM, 88/100
For a small app with actual users, Turso is the strongest $0 in the managed table: the free caps are orders of magnitude past the workload, and the paid step is a published $5.99. Two things to hold: an unindexed SELECT is a full-table bill, and the Supabase join (02.10.2026) is an unanswered ownership question. If local-first reads matter, embedded replicas bolt microsecond reads onto the same account — at 72 in our hub, one notch below the managed row.