Benchmarks

Performance measurements for Raxol's core operations.

Quick start

# Framework comparison (vs Ratatui, Bubble Tea, Textual)
mix run bench/suites/comparison/framework_comparison.exs

# Quick mode (~30s instead of ~2min)
mix run bench/suites/comparison/framework_comparison.exs -- --quick

# Internal benchmarks
mix raxol.bench                    # All benchmarks
mix raxol.bench parser             # Parser only
mix raxol.bench rendering          # Rendering only
mix raxol.bench --quick            # Shorter runs

Latest results

Measured 2026-08-07 at fce2465bb on Apple M1, Elixir 1.20.2 / OTP 29.0.3, with mix run bench/suites/comparison/framework_comparison.exs (full mode). Every number below comes from that one run on that one machine.

Core operations

The emulator ingest rows measure the whole terminal path: parsing plus state application (buffer writes, cursor movement, style changes). The standalone ANSI lexer (mix raxol.bench parser) handles a short plain string in under a microsecond; the two paths differ by orders of magnitude, so any "ANSI parse" number is meaningless without naming which path it measures.

Frame budget

The memory figure is erts_debug.size on the buffer term; cells are stored in a map populated on write, so an empty buffer is small and memory grows with written cells, not dimensions.

Cross-framework comparison

All values in microseconds unless noted. Lower is better.

Raxol: measured on this hardware. Others: published/estimated numbers from different hardware, so the comparison is approximate at best; a same-machine harness is the only honest upgrade. The previous ANSI parse row is gone: published parser numbers for other frameworks measure lexers, while our comparison suite measures full emulator ingest, and comparing the two mislead more than it informed.

Interpretation

Raxol's per-operation latency is higher than Rust/Go (expected for a managed runtime), but:

• Full frame at 5.0ms leaves 69% of the 60fps budget for application logic
• Tree diff at 32us sits between Ratatui (~5us) and Textual (~100us)
• 718K cell writes/sec is more than enough for any terminal UI
• OTP benefits: crash isolation, hot reload, and distribution come built in to the runtime

The BEAM is fast enough for 60fps terminal rendering while also providing fault tolerance and distribution primitives that would require significant additional infrastructure in compiled languages.

Suites

Running benchmarks

# Specific suite files
mix run bench/suites/parser/parser_benchmark.exs
mix run bench/suites/terminal/buffer_benchmark.exs
mix run bench/suites/rendering/render_performance_simple.exs

# Via mix task (uses Benchee, generates HTML reports)
mix raxol.bench parser --dashboard
mix raxol.bench --regression    # Check for regressions (5% threshold)
mix raxol.bench --compare       # Compare with previous run

Performance targets

Tips

• Close other apps for consistent results
• Use --quick for development, full runs for publishing
• Run 3+ times and take the median
• Compare on the same hardware/OS