Rsi Eval

Rsi Eval is a lightweight Python software development kit and command-line tool for evaluating recursive self-improvement methods across agent benchmarks.

RSI Eval is a Python software development kit and command-line interface designed to evaluate recursive self-improvement methods across various agent benchmarks. Users bring their own improvement loop, select tasks, and evaluate candidates through a single interface without requiring optimizer base classes. The catalog includes twenty-four evaluation configurations covering coding, software engineering, reasoning, mathematics, reading comprehension, and domain workflows. It records per-task scores, execution status, token usage, verifier output, and trajectory paths in a common result format. The core software development kit has no third-party runtime dependencies, while execution backends and method dependencies install when needed.

Rsi Eval at a glance

Best for: Software engineer, Researcher
Runs on: Python SDK, CLI
Pricing: Free
Released: October 2026
Category: AI research tool
Website: github.com/danielxu0208/rsi_eval

Key features

Evaluate candidates through a single interface
Reuse benchmark protocols and graders
Read per-task scores and token usage
Record and resume experiments
Browse bundled benchmark configurations
Run optional method workflows

Use it for

Evaluate recursive self-improvement methods
Run agent benchmarks and graders
Record and resume evaluation experiments

Alternatives to Rsi Eval

RSI Anything — It uses world-model predictions and feedback for recursive self-improvement across domains.
Eval Skills — Guides coding agents from first execution to validated graders and experiments.
SoL Pi — Saves $8.75 to $13.50 per hour compared to native Codex and Claude Code systems.
Evals Skills — It helps avoid common mistakes and pitfalls when building AI evaluation suites.
EvalScope — It combines model capability benchmarks with inference performance testing in one tool.
Self Bench — Evaluates coding agents on internal repository tasks instead of generic public datasets.
OpenResearch — Run parallel research workflows locally or on remote compute.
Rigmark — Measures local AI performance with reproducible, agent-like workloads.
Paper2Agent — It enables direct interaction with research papers and automatic result reproduction.
DeepEval — Provides over fifty research-backed metrics and pytest-native execution.
LMMs-Eval — Provides open research and tooling for multimodal model evaluation and benchmarks.
GEAR — Organizes recursive self-improvement through evolutionary search and reproducible evaluation.

All Rsi Eval alternatives