WEEK OF JANUARY 20, 2026

Training Run

Your Weekly AI Conditioning

The authoritative, independent scoreboard for AI model performance. Transparent methodology. Cited sources. No hype—just data.

Our Methodology View Current Scores

Why Training Run Exists

AI is moving fast. Too fast for most people to keep up. Every week brings new models, new benchmarks, new claims. How do you know what's real?

Training Run cuts through the noise. We aggregate data from the most respected evaluation platforms, apply a transparent scoring methodology, and deliver a clear picture of where AI models actually stand—updated weekly.

Whether you're an investor evaluating AI companies, a policymaker shaping regulation, a developer choosing which API to use, or just someone who wants to understand this technology—we're building the resource you need.

The TRS Score

Our Training Run Score (TRS) is a weighted composite of six performance dimensions, aggregated from peer-reviewed benchmarks and public evaluation platforms.

25%

Reasoning

Complex problem-solving and logical deduction

25%

Coding

Real-world programming ability

20%

Human Preference

What users actually prefer in blind tests

15%

Knowledge

Breadth and accuracy across domains

10%

Efficiency

Performance relative to cost

5%

Safety

Reliability and responsible behavior

Learn More About Our Methodology →

Built on Trusted Sources

Every score we publish is derived from publicly verifiable data. No black boxes.

LMSYS Chatbot Arena

1M+ human preference votes via blind comparisons

lmsys.org

ARC-AGI-2

Novel reasoning tasks that test genuine understanding

arcprize.org

SWE-Bench

Real GitHub issues to measure coding ability

swebench.com

MMLU / GPQA

Graduate-level knowledge evaluation

paperswithcode.com

What This Isn't

These aren't predictions about AGI timelines. That's not our department. These are measurements of specific capabilities, not prophecies about AI's future. The trend line is up. What that means? Above our pay grade.

We measure what AI can do today. We cite our sources. We show our math. That's it.

Watch the Show

New episodes every week breaking down the latest in AI performance.

Coming Soon