Hopper

Hopper: Fast Inference for Voice

Hopper is an inference platform designed for voice applications. It trains Speech-to-Text (STT), Text-to-Speech (TTS), and speech Large Language Models (LLMs) on production calls, and then serves and improves them against live traffic.

Key Features

• Voice Model Training and Serving: Trains STT, TTS, and speech LLMs on production calls and improves them using live traffic.
• High-Speed Inference: Achieves 80 ms Time To First Token (TTFT) compared to 600 ms for GPT-4.1.
• Cost-Effective Processing: Priced at $0.50 per 1 million input tokens, compared to $2 for GPT-4.1.
• Model Support: Utilizes the Gemma 4 31B LLM.

Company Background

Hopper is based in San Francisco, CA, and was founded by Pavan Katta and Jashwanth Pedapudi. The team is backed by Y Combinator and industry leaders from Turing, xAI, Stripe, and Rubrik.

Getting Started

Website: https://withhopper.com

Hopper provides fast, cost-effective voice inference infrastructure, enabling teams to optimize and serve STT, TTS, and speech LLMs directly on their live production traffic.