About

Mission

BLXBench provides independent, reproducible benchmarks for AI models. We believe in:

• Transparency — All tests are open source
• Reproducibility — Anyone can run the same benchmarks
• Fairness — No provider pays for placement

What We Do

BLXBench evaluates AI models across focused fixture categories:

1. Speed — How fast does the model respond?
2. Security — Does it refuse harmful requests appropriately?
3. Reasoning — Can it handle complex logical tasks?
4. Debugging — Can it identify and fix bugs?
5. Refactoring — Can it improve code while preserving behavior?
6. Hallucination — Does it stay grounded under tricky prompts?
7. Coding UI — Can it generate and validate UI artifacts?

How It Works

Running Benchmarks

Anyone can run benchmarks using blxbench (install from npm as @bitslix/blxbench — see Installation):

blxbench --headless --provider opr --models openai/gpt-5.4-mini

Results can be submitted to appear on the public leaderboard.

Scoring

Models are scored on:

• Earned score vs max score per test, rolled up to categories and overall
• Rankings use that aggregate; categories with more tests contribute proportionally more to the headline percentage

No Paid Placements

BLXBench does not accept payment for leaderboard placement. Results are based purely on benchmark performance.

Team

BLXBench is operated by bitslix.com.

Contact

See Support for help.