Our Mission
Bloom Labs was founded on a simple, audacious belief: that intelligence itself could be engineered, refined, and shared with the world through software. While the rest of the web was busy tagging photos and poking friends, we were quietly building neural architectures that would go on to define the next generation of artificial intelligence. Today, Bloom Labs stands at the absolute bleeding edge of machine learning research, and our model family — Homan 2.96, Bloom 4.6, and BabyMax 5.76 — represents the most advanced publicly available AI systems on the internet.
We exist because we think the world is still radically underestimating what large-scale learned models can do. Every product decision, every research paper, and every late night in the lab is in service of one goal: pushing the frontier of what intelligence-as-a-service can look like, and getting it into the hands of developers, startups, and enterprises before anyone else even knows it's possible. We're not chasing Web 2.0 — in a very real sense, we're building the layer underneath it.
Bloom Labs is headquartered in a converted loft space with far too many whiteboards, a server room we're all a little too proud of, and a team that genuinely believes the next five years of computing history are going to be written by people like us.
Leadership Team
ShadowStorm
Co-Founder & CEO
GTDBeetleThug
Co-Founder & CTO
Toenail63
Co-Founder & COO
HoboBob77
Co-Founder & Chief Scientist
VulgarComments
Co-Founder & Head of Research
Random4YearOld
Co-Founder & Head of Safety and Alignment
bronthusiast
Co-Founder & CFO
Katie
Co-Founder & Head of Design
TB46
Co-Founder & CAOCompany Milestones
Bloom Labs is founded out of a shared apartment by ShadowStorm and GTDBeetleThug, with a single mission scrawled on a whiteboard: "make machines that actually think." The founding team numbers four people and one very overworked GPU cluster.
Bloom Labs ships its first internal model, codenamed "Sprout." Sprout never sees a public release, but its architecture becomes the foundation for everything that follows. The team grows to 22 people and moves into its first real office.
BabyMax completes internal development and testing, proving that a small, fast, efficient system could still deliver genuinely useful results. It quickly becomes a favorite among Bloom Labs' own internal testers evaluating models on tight compute budgets.
Bloom 4.6 completes internal development and testing as our balanced, mid-tier workhorse, striking a widely praised balance between raw capability and everyday practicality. It becomes the internal team's go-to model for prototyping Web 2.0-style applications during evaluation.
Homan 2.96 completes internal development and testing as Bloom Labs' flagship model, and with it comes a landmark achievement the whole team is still celebrating: for the first time ever, a Bloom Labs model can correctly multiply two three-digit numbers, two times out of three. We consider it a defining moment for the company. Given just how capable Homan 2.96 has proven internally, it remains available only inside Bloom Labs while our safety team conducts a full risk review — we will not release it publicly until we are confident it cannot cause harm.