Is Ultramarine 1 Flash state of the art?Pending benchmarks.Announced Oct 14, 2026; no independent results yet.
Paper Instruments' agentic model for professional knowledge work in finance, law, consulting and operations, post-trained from GLM-5.3 Flash with large-scale reinforcement learning. The API is live; the lab says weights are coming soon.
The details so far
- Lab
- Paper Instruments
- Announced
- Oct 14, 2026
- API
- Live · ultramarine-1-flash
- Open weights
- Announced, not yet published
- Parameters
- ~18B activeunsure
- Base model
- GLM-5.3 Flash
- Focus
- Knowledge work
- Reasoning
- Always on (max by default)unsure · press
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
+4.1 points on average over GLM-5.3 Flash across six knowledge-work benchmarks, with 6.5% fewer tokens.
59.3 on its internal PI Knowledge Work suite at about $0.13 per task.
Sources
LABUltramarine 1 Flash: Scaling Non-Verifiable RLpaperinstruments.com/blog/ultramarine · Oct 14, 2026LABUltramarine 1 Flash product pagepaperinstruments.com/ultramarineCompare when it lands
Unscored: its base model, GLM-5.3 Flash. A scored comparison appears once both are on a shared board.
Tracking continues until an independent board lists it.