Models

Sable 4 is the most capable open-weight family from Oriolis, distilled from the Sable Ultra research line to widen reasoning depth per token.

Vellum 3 Super is a sparse mixture-of-experts model tuned for multi-agent orchestration, long-horizon planning and strict schema output.

Koto K2.5 is a frontier-scale open model with a 245k context window, multi-turn tool calling, vision input and a rewritten tokenizer.

Zenith 4.7 Flash is a low-latency multilingual generation model with a 131,072 token context window and aggressive KV-cache reuse.

Prism 2 [nova] is a 9 billion parameter renderer that turns text prompts into layered compositions and supports region-level editing.

Prism 2 [nova] 4B is an ultra-fast distilled image model. It unifies generation and inpainting in a single checkpoint under 6 GB.

Prism 2 Studio is the production checkpoint of the Prism family, tuned for typography fidelity and repeatable brand palettes.

Cadence 2 (Castilian) delivers streaming speech at 68 ms first-byte latency with prosody control tags and eleven stock voices.

Cadence 2 (English) shares the Castilian acoustic backbone and adds word-level timestamps for caption alignment pipelines.