Inference Frontier

The builders pushing inference forward and sharing how.

A year-round spotlight program recognizing teams shipping real production inference systems. We surface the architecture, tradeoffs, and lessons behind what actually works at scale.

Nominate a team Stay in touch

What this is

A recognition program for teams that shipped real inference systems, learned something meaningful, and can teach the community what good looks like.

We reward systems, not slogans.

What this is not

Not a generic AI innovation award
Not a slide deck competition
Not marketing-driven storytelling
Not a Nebius customer program

Who it’s for

Builders using inference in real systems.

Solo developers and open-source maintainers

Startups and SaaS companies

Enterprise ML platform and infrastructure teams

Researchers building serving or optimization systems

How it works

Nominate a team or apply directly

We perform a short technical sanity review

Selected teams are featured in monthly spotlights

All featured teams are eligible for the end-of-year finals showcase

What you get

Recognition and amplification inside the inference community.

Technical spotlight

Short feature focused on architecture and lessons.

Distribution

Social amplification across Nebius channels.

Stage time

Invitation to speak at Builder Hours, meetups, or partner events.

Peer network

Join a small cohort of production inference builders.

Product feedback loop

Direct line to Nebius product and engineering teams.

Finals showcase

End-of-year recognition moment and awards.

Program cadence

Launch moment

Announce the first cohort (3–5 teams)

Monthly spotlight

1–2 teams per month

Finals night

End-of-year showcase and awards

What we’re looking for

What did you ship?

Real workload. Real users. Real traffic.

What changed?

A concrete architectural or modeling decision.

What improved?

At least one measurable result.

What did you learn?

A pattern others can reuse.

Proof matters

Ranges and deltas are fine. We care about learning, not disclosure. Examples of acceptable evidence:

Tokens per second or throughput at concurrency
p95 / p99 latency or TTFT
KV-cache hit rate improvements
Cost per 1M tokens or cost per request
Error rate reduction under load

Recognition categories

OSS & Independent Builders

Maintainers and solo engineers shipping real inference systems.

Startup Scalers

Small teams running production workloads with measurable growth.

Enterprise Transformers

Platform and infrastructure teams delivering organization-wide inference impact.

Inference Innovators

Breakthroughs in serving, optimization, routing, or post-training systems.

Judging committee

George Cameron

Co-founder at Artificial Analysis

Olga Megorskaya

CEO at Toloka

Laurelle Roseman

Vice-President of Global Partnerships at Nebius

Know a team pushing inference forward?

Nominate a team Stay in touch

Apply with your team or nominate a team

FAQ

No. It is infrastructure-agnostic. Great work counts regardless of where you run today.