LPU-based inference — sub-100ms first-token latencies on open-weight models.
Groq's LPU architecture delivers some of the lowest first-token latencies in the industry across Llama, GPT-OSS, Whisper, and Qwen models. US-hosted. The fast-path route for latency-sensitive applications — chat, agents, interactive tool use — where time-to-first-token is the critical metric.
Every model we route through Groq. Compare residency, ZDR, training posture, and price at a glance — full data-handling detail per route below.
| Model | Region | Zero data retention | Training | Context | Input | Output |
|---|---|---|---|---|---|---|
| US | Not by default | No | 131K | $0.80 | $4.00 | |
| US | Not by default | No | 128K | $0.15 | $0.60 | |
| US | Not by default | No | 128K | $0.07 | $0.30 | |
| US | Not by default | No | — | $0.0019 / min | ||
| US | Not by default | No | — | $0.0007 / min | ||
| US | Not by default | No | 128K | $0.07 | $0.30 | |
| US | Not by default | No | 128K | $0.59 | $0.79 | |
| US | Not by default | No | 128K | $0.05 | $0.08 | |
Groq hosts on 1 route. Each route has its own privacy posture, residency, and GDPR terms. Postures are maintained by Opper with a last-verification timestamp.
Zero data retention is not on by default — abuse monitoring, 30-day retention. No training on customer data. US; SCCs; DPA available.
