---
title: "Introducing LFORLA V2"
description: "A complete rebuild of the LFORLA benchmarking platform — faster runners, private benchmarks and an API-first future."
updated: 2026-03-12
url: https://lforla.org/blog/introducing-lforla-v2
---
We're excited to announce **LFORLA V2** — a complete rebuild of our benchmarking platform.

## What changed

LFORLA V2 replaces the original prototype with a production-oriented stack:

- **Faster evaluation runners.** Benchmark runs are now processed asynchronously and stream results back as they complete, instead of blocking your terminal.
- **Private benchmarks.** Organizations can create benchmarks whose datasets and results never leave the team. Visibility is enforced end-to-end: API, leaderboard and detail views.
- **A cleaner public API.** The V1 API is now the single source of truth for programmatic access to benchmarks, models and the leaderboard.
- **Row-level security everywhere.** Every record is guarded by row-level security policies, so tenant isolation is enforced by the database itself.

## What stays the same

The mission hasn't changed: LFORLA remains the open place to compare reinforcement learning agents across environments and benchmarks, with reproducible scores and transparent methodology.

Head to [the benchmarks](/benchmarks) to see what's live, or read our [methodology guide](/blog/how-we-benchmark-rl-agents) to understand how scores are computed.