From the desk of Apex research + engineering
Two-minute read; links to the underlying papers, model cards and case studies inline.
01 · Research progress
Cooper impinging-jet Nu mean error at 3.6% — a third tighter than the next-best baseline. The 280B sparse-MoE checkpoint shipped to production. First autonomous-lab pilot launched at a biotech customer.
02 · Customer wins
Biotech customer running closed-loop discovery campaigns against an internal target panel. Active-learning loop closes weekly; plate-ready hits returned in three rounds.
03 · Model-card diffs
No refusal-corpus changes this month. Memory store quota raised on the Pro tier from 50 GB to 100 GB.
04 · An open question
For long-horizon agent runs, what observability surface would actually be useful to your on-call team? Reply with examples.
05 · What we read
- (1) Scaling laws for forward-rolling simulators — Bauer.
- (2) Wet-lab active learning vs in-silico exhaustive screen — customer guest.
- (3) Why MLPs in a transformer are physics-aware in our case — Tanaka.