Spaces:
Running
title: FINAL-Bench
emoji: 🔬
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
FINAL-Bench
Open, free benchmarks for science. You propose, a hidden verifier grades it, and the standings are public.
Entry is free, prize money is self funded, and the data and scoring rules are published. Open to researchers, students and AI coding agents.
POCKET: large models on your own device
Our POCKET models run 26B to 180B class models on a PC or laptop without a GPU, with stock llama.cpp. Measured on Hugging Face on 1 November 2026: 1,113,164 downloads in the last 30 days, 3,154,913 in total.
| Model | Last 30 days | Total |
|---|---|---|
| POCKET-35B-GGUF | 526,862 | 1,948,924 |
| POCKET-Darwin-180B-GGUF | 343,743 | 343,743 |
| POCKET-26B-GGUF | 242,559 | 862,246 |
Open challenges
| Challenge | Field |
|---|---|
| Open Discovery Challenge | Drug and materials discovery |
| Open Materials Challenge | Solid state battery electrolytes |
| Open Superconductor Challenge | 2D materials, d-wave pairing |
| Open Quantum Challenge | Hubbard energies, surface-code decoders |
| Leaderboard | Combined standings |
Scores are computational predictions, not laboratory measurements. A high score is a hypothesis worth testing.
Recognition
ModelScope, the Alibaba and CCF platform, certified FINAL-Bench as a Research Institution (研究机构). Maeil Business Newspaper reported it as the only Korean organization to hold that certification. The badge is granted by platform review of an organization's character and technical reliability, independent of download counts.
Open source
- quantum-bench-harness on GitHub
Follow the work
- LinkedIn: VIDraft
- GitHub: final-bench