Groq Removes Multiple Models While Adding High-Performance GPT OSS 20B and 120B with New Prompt Caching Feature

Gemma 2 9B 8k model removed (previously $0.20 input/$0.20 output per million tokens)

Llama Guard 3 8B 8k model removed (previously $0.20 input/$0.20 output per million tokens)

GPT OSS 20B 128k model added at $0.10 input/$0.50 output per million tokens with 1,000 TPS

Kimi K2-0905 1T 256k model added at $1.00 input/$3.00 output per million tokens with 200 TPS

GPT OSS 120B 128k model added at $0.15 input/$0.75 output per million tokens with 500 TPS

DeepSeek R1 Distill Llama 70B 128k model removed (previously $0.75 input/$0.99 output per million tokens)

Prompt Caching feature added with 50% discount on cached input tokens for select models (Kimi K2, GPT OSS 20B, GPT OSS 120B)

Mistral Saba 24B 32k model removed (previously $0.79 input/$0.79 output per million tokens)

Llama 3 8B 8k model removed (previously $0.05 input/$0.08 output per million tokens)

Distil-Whisper ASR model removed (previously $0.02 per hour transcribed with 250x speed factor)

Llama 3 70B 8k model removed (previously $0.59 input/$0.79 output per million tokens)

Batch API discount increased from 25% to 50%

Qwen QwQ 32B (Preview) 128k model removed (previously $0.29 input/$0.39 output per million tokens)