Horizon Labs

Introducing Velocity V1

Project Aurora is a next-generation AI prototype engineered to redefine the boundaries of intelligent systems. Now Supercharged by Velocity V1, Aurora responds up to 10x faster than our previous Aurora 2.1 Snap Model.

Launch Aurora
10×
Faster Inference
99.4%
Accuracy Rate
2.1B
Parameters
47ms
Avg. Latency
780B
Training Tokens
Performance

Benchmarks That
Speak for Themselves.

Velocity V1 was stress-tested across 14 evaluation dimensions — adversarial prompting, long-context retrieval, code synthesis, and multimodal reasoning. The results set a new bar for 2B-scale models.

Reasoning Tasks94.2%
Code Generation91.5%
Multimodal Understanding87.8%
Long-Context Recall89.3%
Safety Evaluations99.1%
Adversarial Robustness96.7%
Roadmap

What's
Coming Next.

With the release of Velocity V1, the next 12 months will see a step-change in what the model can do — and where it can run.

Q1 2026 — Shipped
Aurora 2 Snap
2.1B sparse MoE, 47ms latency, zero-retention inference. Live now.
Q2 2025 — Shipped
Velocity V1
Full-document reasoning and codebase-level analysis without degradation; uses LPU for superior speed.
Q3 2025 — Planned
Velocity V2: Voice & Vision Native
Real-time audio and video frame analysis built into the core model — not bolted on.
Q4 2025 — Planned
On-Device Inference
A quantized 380M distilled model that runs locally. All the reasoning, no cloud dependency.

"The most capable AI should also be the most private. We're proving those two things aren't a tradeoff."

- John Xu · 2026