Homan 2.96 Benchmark Report: Reasoning, Code, and the Curious Case of Multiplication
We are beyond proud to report Homan 2.96's full benchmark results -- including a landmark first for Bloom Labs: correctly multiplying two 3-digit numbers, two times out of three. Full results inside.
Read more »Introducing Homan 2.96: Our Most Capable Model Yet
Today we're proud to announce Homan 2.96, the new flagship of the Bloom Labs model family and the most capable system we've ever internally developed.
Read more »Constitutional Alignment: How We Keep Bloom Labs Models Safe
A look inside the alignment research powering every Bloom Labs release -- how we train our models to be helpful, honest, and firmly on the user's side.
Read more »BabyMax 5.76: Big Intelligence in a Small Package
Meet BabyMax 5.76, our new lightweight model built for speed and scale without sacrificing the smarts.
Read more »Scaling Laws Revisited: What We Learned Training Bloom 4.6
Compute-optimal training notes and lessons learned while building Bloom 4.6, our balanced mid-tier model.
Read more »