The PHP Bench
PHP ORMs 2026
Comprehensive ORM benchmarks for every scenario.
Imports
1
Inbox Rows
1
Runs
312
Metrics
16,224
Matrices
52
Scenarios
6
Fastest Stack
Atlas
Storage
all-engines
Speed
0.060988s
Fastest Combinations
Inbox
| ID | Status | Runner | Runtime | File |
|---|---|---|---|---|
| #1 | processed | runner-20260329222123 |
39.501s avg 0.126606s |
results-20260329203643-74505162-default-20260329-203643.csv |
How Scenario Families Shift Rankings
| Family | Pressure | Why It Changes Speed | Avg Runtime |
|---|---|---|---|
| simple | baseline row shape | Shows framework overhead with very little field-mapping noise, so identity-map, query builder, and per-record object cost dominate. | 0.099689s |
| canary | mixed nullable and cast-heavy fields | Wide records amplify hydration, casting, dirty-checking, and serialization overhead. Frameworks with expensive attribute normalization usually spread out here. | 0.154049s |
| int_fixed | narrow scalar writes | A near-floor for typed persistence. If an ORM is still slow here, the cost is usually framework machinery rather than payload complexity. | 0.123969s |
| float_fixed | numeric conversion and precision paths | Decimal and floating-point mapping can trigger extra normalization work, especially on update verification and dirty-state comparison. | 0.141195s |
| string_fixed | predictable text hydration | Fixed-length text stresses text field assignment without the variability of large blobs, so column-count and hydration strategy matter more than payload churn. | 0.118621s |
| string_variable | application-style text records | This is the closest to a real CRUD app payload. Variable strings and notes fields widen the gap between lightweight row mappers and heavier active-record layers. | 0.122112s |
| Operation Shape | Why Models Move |
|---|---|
| Warmup vs hot runs | Run 1 is a warmup pass. Hot-path rankings exclude it so adapter boot, metadata discovery, and first-query setup do not drown out steady-state behavior. |
| Insert and indexed schemas | Indexed variants make inserts more expensive because each row write updates secondary indexes. ORMs that already have heavy unit-of-work overhead get exposed harder there. |
| Read all vs read probes | Read All rewards efficient bulk hydration. Random and tail probes reveal per-query setup cost, identity lookup behavior, and how much work happens for tiny result sets. |
| Update with verification | Updates are followed by read-back verification. That means slow dirty-checking, full-row writes, cast normalization, and post-update reads all feed into the total. |
| Delete phase | Delete cost is not just delete. The harness verifies the row is actually gone, so frameworks with slow teardown or follow-up lookup behavior pay twice. |
| Total runtime | Total runtime is a whole-machine workload number: init, insert, every read shape, every update shape, verification passes, and delete. |
Storage Engine Curve
sqlite-memory
sqlite-memory0.069814s
mariadb
mariadb0.127033s
sqlite-file
0.142422s
Scenario Families
Imports
| Run | Host | Status | Rows | Runtime | Started |
|---|---|---|---|---|---|
| 20260329203643-74505162 | eimu 8.5.3 |
complete | 16,224 | 39.501s avg 0.126606s |