koral
Menu

Benchmarks / Recorded evidence

Measure the program.
Keep the conditions.

A recorded Counter comparison with matching inputs and final state. Read the scope before reading the numbers.

One Counter. Three implementations.

Native Rust, Pinocchio, and Koral were compared in a local LiteSVM environment. These are recorded measurements, not live browser results. Lower compute units mean less computation for the specific instruction.

Recorded November 11, 2026 · Compute units (CU)
InstructionNative RustPinocchioKoral
initialize4,6013,1863,366
increment857113107

What was held constant?

  • The same inputs and account addresses.
  • One initialization and 20 increments.
  • A matching final 49-byte account state.
  • LiteSVM 1.6.1 and Agave 3.1.10 / platform-tools 1.52.

What do the numbers say?

Koral initialization currently uses more CU than Pinocchio. Its recorded increment is slightly lower in this Counter scenario. A small instruction difference does not establish a general compiler advantage.

CU describes computation. Account storage deposits are a separate cost. These measurements do not describe storage rent, complete DEX execution, security, or real-world throughput.

Inspect the snapshot.

The downloadable JSON records the environment, call count, final-state agreement, and values shown above. Full reproduction requires the development utility checkout; a public source release is pending.

Download measurement snapshot ↗