Skip to content

Rankings Industries Companies Research Blog

Sign in List your company

ResearchFormerly WizWand

Papers Benchmarks Datasets Tasks

Research
Benchmarks
Downstream accuracy extrapolation on PIQA

Downstream accuracy extrapolation on PIQA

0.025RMSECE Loss

Updated 5mo ago

Evaluation results

Method	Links
CE Loss2026.05		0.025
Proxy2026.05		0.029
Compute2026.05		0.068

State-of-the-art paper

CE Loss

Language models scale reliably with over-training and on downstream tasks

Dataset

PIQA

Follow for updates

Join SOTA2 Contact

Related benchmarks

Model Ranking Prediction on Reasoning Tasks Aggregate Model Ranking Prediction on AIME Model Ranking Prediction on GPQA Model Ranking Prediction on HMMT Model Ranking Prediction on MMLU-Pro

Building an AI company?

Claim your profile, keep your facts current and see how buyers find you. Rankings are never paid for.

How rankings work List your company

Find AI companies and products for your business problem. Compare solutions and evidence by industry and task.

Explore

Rankings
Industries
Categories
Companies
Top AI companies

Read

Blog
Research
llms.txt

For AI companies

List your company
Claim a profile
Ranking badges

SOTA2 with AI

Claude connector
ChatGPT plugin
Tool reference
Connector privacy
Connector terms

SOTA2

Why SOTA2
How we rank
Contact

© 2026 SOTA2Built in Bellevue, USA