Model Recipes
Curated Qwen3.5, Gemma 4, and DeepSeek V4 Flash recipes with pinned revisions.
Qwen Starter
Compact Qwen3.5 recipes for dev iteration, agent prototyping, and the Toolkit-Chat dogfood path.
Qwen Pro
Mid-size Qwen3.5 for production inference and fine-tuning on isolated capacity.
Qwen Max
Largest Qwen3.5 for heavy inference and training on dedicated B200/H200 hardware.
Gemma Lite
Google Gemma 4 E4B for lightweight inference and rapid prototyping.
Gemma Vision
Gemma 4 26B-A4B with vision capabilities for multimodal workloads.
Gemma Max
Largest Gemma 4 at 31B for demanding training and inference tasks.
DeepSeek Flash
DeepSeek V4 Flash for high-throughput serving on dedicated Blackwell hardware.