# How Paper Instruments uses Fireworks AI

**~98%**: “Outscoring Fable 5 at ~98% lower inference cost”

As published on [fireworks.ai](https://fireworks.ai/customers/paper-instruments-trains-a-frontier-level-model-for-knowledge-work-at-98-percent-lower-cost), October 2026. Captured by usedby on 2026-10-31.

- Company: [Paper Instruments](https://www.usedby.ai/companies/paper-instruments.md)
- Tool: [Fireworks AI](https://www.usedby.ai/tools/fireworks-ai.md)
- Teams: Research

## The use case: how Paper Instruments deploys Fireworks AI

Paper Instruments trains its Ultramarine 1 Flash model, an open model for professional analysis, writing, and presentations, using reinforcement learning on the Fireworks Training API and Fireworks inference. Fireworks provides rollout generation, GPU tuning, and training/inference numerical alignment, and serves the resulting model through serverless inference.

Summary written by usedby from the source page, in English. The figures are those of Fireworks AI and Paper Instruments, not ours.

> On Paper Instruments’ internal PI Knowledge Work benchmark, Ultramarine 1 Flash scored 59.3 versus Claude Fable 5’s 57.8, at $0.13 versus $8.19 per task.

- **4.1 points**: “Ultramarine improved on GLM-5.3 Flash across all six reported knowledge-work benchmarks, gaining 4.1 points on average.”
- **7.3-point**: “On Paper Instruments’ internal PI Knowledge Work benchmark, Ultramarine’s score rose from 52.0 to 59.3, a 7.3-point improvement.”
- **$300,000**: “Total development cost was approximately $300,000, including the data creation and experimentation that preceded that run.”
- **two weeks**: “After developing its curriculum through smaller-model experiments, the team completed the GLM-5.3 Flash training job on Fireworks in approximately two weeks.”

> We retained full control over sampling and shaping the reward. We could focus on the RL science without having to think about kernel-level optimization, sizing, and parallelism.
>
> Daanish Khazi, Co-founder

Source: [fireworks.ai](https://fireworks.ai/customers/paper-instruments-trains-a-frontier-level-model-for-knowledge-work-at-98-percent-lower-cost), captured 2026-10-31.

## Other AI tools used by Paper Instruments

- [E2B](https://www.usedby.ai/tools/e2b.md) — Observed

## Other companies using Fireworks AI

- [Cursor](https://www.usedby.ai/companies/cursor.md) — Confirmed
- [Genspark](https://www.usedby.ai/companies/genspark.md) — Confirmed
- [Notion](https://www.usedby.ai/companies/notion.md) — Confirmed
- [Phylo](https://www.usedby.ai/companies/phylo.md) — Confirmed
- [Trilogy](https://www.usedby.ai/companies/trilogy.md) — Confirmed
- [Upwork](https://www.usedby.ai/companies/upwork.md) — Confirmed

## What usedby checked

We compared the story with its live page on 2026-10-31.

- Checked: the figure ~98% is printed word for word on the page, near the name of Paper Instruments.
- Checked: the passage quoted above is copied word for word from the page, near the name of Paper Instruments.
- Checked: each number in our summary is printed on the page.
- Checked: the publication date is read from the page’s own metadata, never guessed.
- Checked: Paper Instruments uses Fireworks AI. Confirmed line. Latest check across sources: 2026-10-31.
- Not checked: the result itself. We quote it; we did not measure it.

---
Source: https://www.usedby.ai/case-studies/paper-instruments-fireworks-ai · How we check: https://www.usedby.ai/methodology