Pull any model by name — each ships as a single verified .oasr bundle with weights, tokenizer, and metadata baked in. Public listings are generated from the release catalog, with real hashes, sizes, licenses, and measured metrics only.
qwen3-asr-1.7b Recommended Multilingual · 3.9 GB peak RAM · Apache-2.0 xasr-zh-en Recommended Multilingual · 1.3 GB peak RAM · Apache-2.0 qwen3-asr-0.6b Recommended Multilingual · 2.7 GB peak RAM · Apache-2.0 cohere-transcribe-03-2026 Multilingual · 3.2 GB peak RAM · Apache-2.0 moonshine-tiny English · 292 MB peak RAM · MIT whisper-base Multilingual · 386.1 MB peak RAM · Apache-2.0 whisper-base.en English · 383.3 MB peak RAM · Apache-2.0 whisper-large-v3 Multilingual · 3.8 GB peak RAM · Apache-2.0 whisper-large-v3-turbo Multilingual · 2.1 GB peak RAM · MIT whisper-medium Multilingual · 2 GB peak RAM · Apache-2.0 whisper-medium.en English · 2 GB peak RAM · Apache-2.0 whisper-small Multilingual · 959.2 MB peak RAM · Apache-2.0 whisper-small.en English · 860.7 MB peak RAM · Apache-2.0 whisper-tiny Multilingual · 262.2 MB peak RAM · Apache-2.0 whisper-tiny.en English · 263.7 MB peak RAM · Apache-2.0
A .oasr bundle is a single verified artifact:
quantized weights plus runtime metadata, pinned by catalog hash for reproducible installs.