K-Dense Updates

// July 2, 2026

>What’s New?

Long time, no talk, K-Dense community! The agentic science space has been anything but quiet this week. From new model launches to louder debates about what “AI for science” should look like in practice, the past few weeks have made one thing clear: the next wave of scientific tooling is coming, and K-Dense is ready for the challenge.

Some highlights over the last month:

•Co-Founder & CTO Timothy Kassis attended the Google DeepMind Startup Forum and participated in two panel discussions. It was energizing to be in rooms where researchers, founders, and operators are thinking seriously about what it takes to turn powerful models into real-world scientific workflows.

•We welcomed a new addition to our staff! Give a warm welcome to our Scientific Communications & Community Lead, Maya Navarro, and follow her @kdensemaya on X.

•Updated the Ginkgo Cloud Lab skill. Check out the full scientific skills library (and maybe become our 30,000th stargazer!).

>K-Dense BYOK vs Claude Science

Our open-source K-Dense BYOK (bring-your-own-key) offers an open-source alternative to Claude Science, providing greater database coverage, support for 200+ scientific file formats, the ability to choose your own model, and add scalable compute resources.

See it in action →

Check out the repo →

>Webinar: Build Your Own Scientific Agent Skills

Missed the 2-hour workshop on skills at The Agentic Scientist Summit? Not to worry! We’re bringing the beginner-friendly intro to agent skills to a virtual-friendly workshop on July 8 at 1:30 PM PST.

K-Dense AI engineer, Yuhuan He, will be giving an encore demonstration of how to install a ready-made skill and then build one live.

Learn more and register →

>Recommended Reading

•We were part of an early access program to try the latest Google models. We benchmarked Nano Banana 2 Lite against leading scientific image models across 240 generated figures with surprising results. We did the same for Omni Flash on scientific video generation. The bottom line: the models aren’t quite ready for primetime when it comes to scientific applications. See the full writeup here.

Nano Banana 2 Lite generation speed benchmark results

•Are AI agent skills just API docs with a fancy name? We put it to the test. NVIDIA provided early access to their BioNeMo Agent Toolkit Skills and we ran 830+ controlled experiments with their skills to test performance. Check out the key takeaways. (Did we mention that we also got a shout-out on stage from NVIDIA’s own VP of Healthcare, Kimberly Powell, at BIO2026 this month?!)

NVIDIA BioNeMo skills benchmark: success rate vs cost per run

•If you haven’t heard enough about benchmarking, check out this thought piece on benchmarking AI co-scientists against real-world discovery.

>Stay Connected

Until next time,

The K-Dense Team

K-Dense Inc.

380 Portage Avenue

Palo Alto, CA 94306

Unsubscribe