All Models

Qwen3.8 27B dense model (August 2026, open weights), improving on 3.6-27B in coding and office tasks; 1M context on Model Studio. The newest dense Qwen model.

Canopy Labs' Orpheus text-to-speech model for the Saudi Arabic dialect, with six voices (Abdullah, Fahad, Sultan, Lulwa, Noura, Aisha) and WAV output; it ignores vocal-direction tags. Groq caps requests at 200 characters and requires accepting Canopy Labs' terms.

Canopy Labs' Orpheus v1 English TTS, which replaced PlayAI voices on Groq at the end of 2025: six voices and bracketed style directions such as [cheerful] or [whisper]. Input is limited to 200 characters per request and output is WAV only.

OpenAI's safety-classification fine-tune of gpt-oss-20b (October 2025, Apache 2.0): it reasons over a policy you write and returns a decision with its reasoning, instead of scoring fixed moderation categories. A 120B sibling also exists.

The smaller gpt-oss open-weight model (August 2025, Apache 2.0) on DeepInfra and Groq: 21B parameters with 3.6B active, adjustable reasoning effort and a 128K context, built for low-latency use.

OpenAI's larger open-weight reasoning model (August 2025, Apache 2.0), served here by DeepInfra, Together and Groq: a 117B MoE with 5.1B active, configurable reasoning effort and a 128K context.

Meta's smallest Prompt Guard 2 classifier (April 2025): a DeBERTa-v3-xsmall model labelling text benign or malicious for prompt injection and jailbreaks, with a 512-token window and about 75% lower latency than the 86M version but weaker non-English accuracy.

The larger Prompt Guard 2 classifier from Meta (April 2025), built on multilingual mDeBERTa-v3 to flag prompt-injection and jailbreak text across languages; Meta reports 97.5% English recall at a 1% false-positive rate. Inputs over 512 tokens must be split.

Whisper large-v3-turbo (October 2024), OpenAI's pruned large-v3 with 4 decoder layers instead of 32 (809M parameters), giving much faster transcription at a small accuracy cost; it does not translate.

Alias for Whisper large-v3-turbo that is fulfilled by Groq, not OpenAI's API: the October 2024 open-weight model decodes several times faster than large-v3 for a little accuracy and handles transcription only.

OpenAI's Whisper large-v3 (November 2023), a 1.55B-parameter multilingual speech model; this version supports both transcription and translation into English, with word and segment timestamps and files up to 100 MB.

Compatibility alias for OpenAI's open-weight Whisper large-v3 that is served by Groq rather than OpenAI's API; the 2023 model recognizes roughly 99 languages and can translate speech into English.