Benchmarks
Rezi includes a benchmark suite comparing terminal UI pipelines across 22 scenarios. The authoritative write-up, methodology, and raw artifacts live in:
Rezi includes a benchmark suite comparing terminal UI pipelines across 22 scenarios. The authoritative write-up, methodology, and raw artifacts live in:
BENCHMARKS.md— full scenario definitions, results, and caveatsbenchmarks/— committed JSON + Markdown result sets
Frameworks
| Framework | Runtime | Notes |
|---|---|---|
| Rezi | Node.js | Full layout engine, binary drawlist, native C renderer |
| Ink | Node.js | React reconciler + Yoga layout + ANSI string output |
| OpenTUI (React) | Bun | React declarative driver on @opentui/core |
| OpenTUI (Core) | Bun | Direct imperative API, no React overhead |
| Bubble Tea | Go binary | Elm architecture, lipgloss string rendering |
| terminal-kit | Node.js | Low-level terminal buffer, no layout system |
| blessed | Node.js | Imperative terminal UI, no constraint layout |
| Ratatui | Rust binary | Native immediate-mode renderer |
Selected results — 2026-02-27, M4 Pro, PTY mode
Single-replicate quick run. Numbers are directional; see BENCHMARKS.md for caveats.
Primitive workloads
rerender — one value changes in a stable tree:
| Framework | Mean | ops/s |
|---|---|---|
| terminal-kit | 60µs | 16,400 |
| Ratatui | 70µs | 14,300 |
| blessed | 71µs | 14,000 |
| Rezi | 391µs | 2,600 |
| OpenTUI (Core) | 1.18ms | 850 |
| OpenTUI (React) | 2.72ms | 368 |
| Bubble Tea | 8.32ms | 120 |
| Ink | 20.03ms | 50 |
tree-construction — build a widget tree from scratch:
| Framework | 100 items | 1000 items |
|---|---|---|
| Rezi | 209µs | 1.55ms |
| terminal-kit | 203µs | 1.91ms |
| Ratatui | 917µs | 2.37ms |
| OpenTUI (Core) | 1.59ms | 17.60ms |
| blessed | 1.82ms | 19.73ms |
| Ink | 24.61ms | 78.94ms |
| OpenTUI (React) | 10.67ms | 102.49ms |
Full-app workloads (PTY output)
terminal-full-ui — composite dashboard, 24 services, 40×120 terminal:
| Framework | Mean | ops/s | Peak RSS |
|---|---|---|---|
| Ratatui | 267µs | 3,700 | 2.8 MB |
| blessed | 331µs | 3,000 | 100 MB |
| Rezi | 1.24ms | 806 | 89 MB |
| OpenTUI (Core) | 1.31ms | 765 | 107 MB |
| OpenTUI (React) | 3.15ms | 317 | 162 MB |
| Bubble Tea | 8.33ms | 120 | ~2 MB |
| Ink | 22.09ms | 45 | 170 MB |
terminal-strict-ui — structured multi-panel layout (header, 3-column body, footer, status bar):
| Framework | Mean | ops/s |
|---|---|---|
| Ratatui | 189µs | 5,300 |
| blessed | 334µs | 3,000 |
| Rezi | 950µs | 1,100 |
| OpenTUI (Core) | 1.27ms | 788 |
| OpenTUI (React) | 4.60ms | 217 |
| Bubble Tea | 8.32ms | 120 |
| Ink | 22.41ms | 45 |
terminal-virtual-list — 100K items, 40-row viewport, full PTY output:
| Framework | Mean | ops/s |
|---|---|---|
| Ratatui | 129µs | 7,700 |
| blessed | 154µs | 6,500 |
| Rezi | 798µs | 1,300 |
| OpenTUI (Core) | 1.09ms | 916 |
| OpenTUI (React) | 6.75ms | 148 |
| Bubble Tea | 8.32ms | 120 |
| Ink | 22.56ms | 44 |
Memory
Peak RSS at terminal-level workloads:
| Framework | Typical range |
|---|---|
| Ratatui | 2–3 MB |
| Bubble Tea | 1–30 MB |
| Rezi | 83–215 MB |
| terminal-kit | 83–142 MB |
| blessed | 85–420 MB |
| OpenTUI (Core) | 85–243 MB |
| Ink | 128–1,250 MB |
| OpenTUI (React) | 96–5,440 MB |
Running benchmarks
Prerequisites
# Build Rezi
npm ci && npm run build && npm run build:native
npx tsc -b packages/bench
# Bun — required for OpenTUI scenarios
# Install: curl -fsSL https://bun.sh/install | bash
# Go — required for Bubble Tea (built automatically on first run)
# Rust/cargo — required for Ratatui (must be built manually):
cd benchmarks/native/ratatui-bench && cargo build --release && cd -Quick run — all frameworks, PTY mode
PATH="$HOME/.cargo/bin:$HOME/.bun/bin:$PATH" \
REZI_BUN_BIN="$HOME/.bun/bin/bun" \
node --expose-gc packages/bench/dist/run.js \
--suite all --io pty --quick \
--output-dir benchmarks/local-allFull run (more iterations)
PATH="$HOME/.cargo/bin:$HOME/.bun/bin:$PATH" \
REZI_BUN_BIN="$HOME/.bun/bin/bun" \
node --expose-gc packages/bench/dist/run.js \
--suite all --io pty \
--output-dir benchmarks/local-fullRigorous terminal suite — multi-replicate, shuffled
Suitable for publication-grade comparisons. Requires bare-metal Linux with CPU pinning for minimal jitter:
node --expose-gc packages/bench/dist/run.js \
--suite terminal --io pty \
--replicates 7 --discard-first-replicate \
--shuffle-framework-order --shuffle-seed my-run \
--cpu-affinity 0-7 --env-check warn \
--output-dir benchmarks/local-terminalTargeted runs
# One scenario across all frameworks
node --expose-gc packages/bench/dist/run.js \
--scenario terminal-full-ui --io pty --quick
# One framework across all scenarios
node --expose-gc packages/bench/dist/run.js \
--framework rezi-native --io pty --quick
# Rezi vs OpenTUI Core
node --expose-gc packages/bench/dist/run.js \
--matchup rezi-opentui --opentui-driver core \
--io pty --quick --output-dir benchmarks/local-matchupKey flags
| Flag | Description |
|---|---|
--suite all|terminal | All scenarios, or terminal-only |
--framework <name> | Single framework |
--scenario <name> | Single scenario |
--io stub|pty | Stub (no PTY) or real terminal write |
--quick | Reduced iteration counts for fast checks |
--replicates <n> | Repeat the full run N times |
--output-dir <path> | Write results.json + results.md |
See BENCHMARKS.md for the full scenario reference, methodology notes, and interpretation guidance.
Constraint Expressions (DSL Reference)
expr("...") is Rezi’s constraint expression escape hatch. It’s used for derived layout relationships (parent/viewport/sibling/intrinsic).
Widget API Reference
Rezi ships a comprehensive built-in widget catalog. Every widget is a plain TypeScript function that returns a VNode -- the virtual-DOM node Rezi reconciles, lays out, and renders to the terminal.