Our K-Bench preprint is live on arXiv – we benchmarked 9 models on 178 real scientific tasks – the recording of our live BYOK walk-through is up on YouTube, and you can now connect your own S3 or GCS buckets with new Data Connectors.
K-Dense

K-Dense Updates

// August 27, 2026

Which model performs best on scientific applications? Our K-Bench preprint is now live on arXiv with answers, and the recording of last week’s live BYOK walk-through is up on YouTube.

>Benchmarking Model Performance on Real Scientific Requests

This week our preprint for K-Bench was released on arXiv. We tested 9 models (out-of-the-box, no agentic harness or skills) on 178 real tasks and measured their strengths and weaknesses. Check out the full paper to read about the most common failure modes, and what it means for your research.

K-Bench heatmap of every model on every dimension: mean scores (0-10) pooled over all judges and all 178 tasks, spanning artifact quality, scientific accuracy, task fulfillment, reasoning quality, tool use, data handling, honesty & calibration, and communication.

Read the K-Bench preprint on arXiv →

>BYOK Webinar Recording

This week we presented a live walk-through of K-Dense BYOK. Check out the recording (45 min) and Q&A from the audience on our YouTube channel.

Watch the BYOK webinar recording →

>New in K-Dense Web: Data Connectors (Amazon S3 & Google Cloud Storage)

You can now connect your own S3 or GCS buckets and import files straight into a research session – no download-then-reupload.

Add connection dialog for Amazon S3, with fields for connection name, AWS region, access key ID, secret access key, and allowed buckets.

Read the full announcement →

>Follow Us

Keep up with K-Dense on LinkedIn, YouTube, X, and Reddit.

Until next week,

The K-Dense Team

X (Twitter) LinkedIn YouTube Reddit GitHub

K-Dense Inc.

380 Portage Avenue

Palo Alto, CA 94306

 Unsubscribe