---
title: "What Is LFORLA? The Open LLM Benchmarking Platform"
description: "LFORLA is an open platform that benchmarks large language models on standardized tests and publishes every score on a free live leaderboard — with cost per task, latency and full metric breakdowns."
updated: 2026-08-21
url: https://lforla.org/blog/what-is-lforla
---
# What Is LFORLA? The Open LLM Benchmarking Platform

**LFORLA is an open, public platform for benchmarking large language models (LLMs). It measures how AI models perform on standardized tests — called benchmarks — and publishes every score on a free, live leaderboard at lforla.org.**

Unlike closed rankings compiled from vendor marketing, every LFORLA score comes from a reproducible evaluation run that anyone can verify. Results are submitted through the open `lforla-eval` CLI, stored in a public database, and displayed with full metric breakdowns: accuracy, cost per task, latency and token usage.

## What does LFORLA measure?

LFORLA evaluates models across several dimensions:

- **Overall score** (0–100): the weighted aggregate of a benchmark's metrics.
- **Per-metric scores**: each benchmark defines its own rubric — role coverage, skill match, budget compliance for recruitment tasks; rubric conformity and completeness for exam-style questions.
- **Cost efficiency**: cost per token, per action and per task, so you can compare not just *which* model is smartest but *which* delivers the most value per dollar.
- **Latency**: average response time per sample.

## Which benchmarks are available?

LFORLA hosts benchmarks across multiple domains:

| Benchmark | Domain | What it tests |
|---|---|---|
| Team Recruitment (Oracle) | Agentic reasoning | Tool use, constraint satisfaction, team building |
| DEC Accounting | Professional knowledge | French accounting expertise, ethics, statutory audit |
| PolitiScales Bias | Social bias | Political leaning detection across 117 statements |
| Root Cause Analysis | Business reasoning | Structured problem diagnosis |
| FISY | Financial modeling | Financial analysis workflows |
| Scroll series | Scientific ML | Ink detection, mesh refinement, surface segmentation |

Each benchmark page publishes its system prompt openly, so researchers know exactly what a model was asked to do.

## How are results verified?

Every submission carries its raw metrics JSON, sample count, token counts and cost data. Submissions start as *pending* and are marked *verified* by moderators once reviewed. The leaderboard recomputes from this public table — there is no manual ranking.

## Key facts about LFORLA

- **Platform type**: open LLM benchmarking and leaderboard platform.
- **Access**: 100% free, no account needed to read results; a free account is needed to submit.
- **Method**: reproducible CLI-driven evaluations (`lforla-eval`), public prompts, per-metric scoring with an LLM-as-judge for open-ended answers.
- **Cost tracking**: native per-task and per-token cost metrics on every run.
- **API**: public REST API under `https://api.lforla.org/api/v1` (health, leaderboard, benchmarks, models, stats).

## Frequently asked questions

**Is LFORLA free?**
Yes. Browsing the leaderboard, benchmarks and API is free. Submitting your own runs requires only a free email-verified account.

**How is LFORLA different from other AI leaderboards?**
Three ways: (1) every score links back to a reproducible run rather than a self-reported claim, (2) cost-per-task is a first-class metric alongside accuracy, and (3) benchmark prompts are published openly instead of kept secret.

**Can I add my own benchmark?**
Yes. Benchmarks are defined in a YAML spec (tasks, metrics, scoring rubric) and pushed with `lforla-eval bench-push`. See the [CLI documentation](/cli).

**Which models are tracked?**
All major frontier and open-weight models: GPT-4o and later OpenAI models, Claude 3 family and later Anthropic models, Gemini 1.5/2.x, DeepSeek V3/R1, and more — see the [model catalog](/models).