Builders & open source
Oleksiy Ostapenko
Core contributor to Super Apriel, ServiceNow Research's 15B supernet with switchable attention mixers
Oleksiy Ostapenko is one of three core contributors to Super Apriel, a 15-billion-parameter supernet from ServiceNow Research in which every decoder layer offers several trained mixer choices, released as SuperApriel-15B-Base and SuperApriel-15B-Instruct. The paper's contributions section credits Ostapenko with model development, supernet training, serving infrastructure and throughput benchmarking. Ostapenko is also first author of the Apriel-H1 technical report (November 2025) on hybrid attention and Mamba models.123
At a glance
- Current role
- None confirmed in the sources
- Sector
- Company
- Work listed
- 1
- Sources cited
- 5
- Profile reviewed
Apriel-H1
Ostapenko is first author of the Apriel-H1 technical report (November 2025) from ServiceNow, which lists Ostapenko among its core contributors. The report introduces hybrid 15B reasoning models obtained by incremental distillation from Apriel-Nemotron-15B-Thinker, replacing less critical attention layers with Mamba blocks. Ostapenko also co-wrote ServiceNow-AI's November 2025 post describing the approach.345
Super Apriel
The Super Apriel paper (April 2026) lists Ostapenko first among three core contributors, alongside Raymond Li and Torsten Scholak, with authors sorted alphabetically. The model is a 15B-parameter supernet whose decoder layers each provide four trained mixer choices, so that one checkpoint can be served at several speeds.12
According to the paper's author contributions, Ostapenko worked on model development and experiment planning, optimized the vLLM supernet serving infrastructure, ran supernet training through distillation and supervised fine-tuning, contributed to mixer initialization and placement evaluation code, and benchmarked inference throughput.2
Work and contributions
SuperApriel-15B-Instruct Open weights
Core contributor to Super Apriel who ran supernet training (distillation and supervised fine-tuning), optimized the vLLM serving infrastructure and benchmarked inference throughput.2
Timeline
Documented roles
The sources do not confirm a current role; these are the roles they document.
- ServiceNow (author affiliation on the November 2025 Apriel-H1 and April 2026 Super Apriel papers), ServiceNow24
Companies in their history
- ServiceNow 6 people in the wiki
Related people
- Shruthan RadhakrishnaConnected to ServiceNow
- Torsten ScholakConnected to ServiceNow
- Raymond LiConnected to ServiceNow
- Sathwik Tejaswi MadhusudhanConnected to ServiceNow
- Paul SmithConnected to ServiceNow
- Song HanInference and serving · Language models
- Arjun DesaiLanguage models · Inference and serving
- Younes BelkadaLanguage models · Inference and serving
- Eric HartfordLanguage models · Inference and serving
- BartowskiInference and serving · Language models
Suggest a correction or removal
Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.
References
5 sources · profile reviewed Oct 22, 2026 · data updated Oct 22, 2026
- Super Apriel: One Checkpoint, Many Speeds (arXiv 2604.19877)
- Super Apriel: One Checkpoint, Many Speeds — full text (authors, affiliations and contributions)
- Apriel-H1: Towards Efficient Enterprise Reasoning Models (arXiv 2511.02651)
- Apriel-H1: Towards Efficient Enterprise Reasoning Models — full text, Contributions and Acknowledgments
- Apriel-H1: The Surprising Key to Distilling Efficient Reasoning Models (ServiceNow-AI on Hugging Face)