Groq LPU Inference API Status, Latency & Hosted LLM Directory (2026)

What is the latency, uptime SLA, and model lineup of Groq LPU Inference?

Groq LPU Inference maintains an average response latency of 58ms TTFT and an operational uptime SLA of 99.98% across its global infrastructure endpoints (us-east-1, us-west-1, eu-west-1). It currently hosts 5 tracked foundation models including Qwen 3.8 Flash Next, Gemma 4 31B IT, Llama 4 Maverick, Llama 3.3 70B Instruct.

Verified daily via automated API latency tests and official documentation.

Zero-Shot RAG Verified

Average TTFT Latency 58ms P90 Benchmark Window

Historical Uptime SLA 99.98% 30-Day Rolling Telemetry

Regional Deployment

us-east-1 us-west-1 eu-west-1

Multi-AZ Availability

Models Hosted on Groq LPU Inference (5)

Qwen 3.8 Flash Next

Alibaba Cloud · 125B/6B Sparse MoE (512 Experts, 10 Active) with 51B N-Gram Embedding & Vision Encoder

85.1 IVR

High-speed Apache 2.0 sparse MoE model with 1M context window and sub-100ms TTFT latency.

Input / Out $0.16 / $0.47

Host TTFT 255ms

Throughput 296 t/s

Gemma 4 31B IT

Google · Dense Open-Weight Transformer

80.3 IVR

Google's flagship Gemma 4 31B dense open-weight model delivering frontier-class coding and reasoning on a single enterprise GPU.

Input / Out $0.12 / $0.36

Host TTFT 75ms

Throughput 279 t/s

Llama 4 Maverick

Meta · Multimodal Sparse MoE (17B Active / 400B Total)

77 IVR

Meta's flagship 128-expert open-weight multimodal MoE model delivering 1M context and rapid LPU/CS-3 inference throughput.

Input / Out $0.20 / $0.60

Host TTFT 80ms

Throughput 420 t/s

Llama 3.3 70B Instruct

Meta · Dense Transformer with GQA

74.6 IVR

Meta's proven 70B dense open-weight model widely deployed across hardware accelerators (Groq, Cerebras) for ultra-fast text generation.

Input / Out $0.12 / $0.30

Host TTFT 75ms

Throughput 363 t/s

DeepSeek R1

DeepSeek · MoE Reasoning Transformer

64.8 IVR

MIT-licensed open-weight reasoning powerhouse trained via large-scale reinforcement learning for mathematical and algorithmic synthesis.

Input / Out $0.55 / $2.19

Host TTFT 850ms

Throughput 150.8 t/s

Frequently Asked Questions About Groq LPU Inference

What is the average latency and uptime SLA of Groq LPU Inference?

Groq LPU Inference delivers an average Time to First Token (TTFT) latency of 58ms and a verified 30-day rolling uptime SLA of 99.98%.

Which AI models are available on Groq LPU Inference?

Groq LPU Inference hosts 5 tracked LLMs on InferenceRate: Qwen 3.8 Flash Next, Gemma 4 31B IT, Llama 4 Maverick, Llama 3.3 70B Instruct, DeepSeek R1.

Which cloud regions does Groq LPU Inference support?

Groq LPU Inference supports API inference deployments across us-east-1, us-west-1, eu-west-1.