DeepSeek R1 distilled on Llama 70B — strong reasoning model.

Gemma 2 9B — Google’s efficient instruction-tuned model.

Gemma 7B — legacy Google model. Prefer gemma2_9b_it.

Llama 3 70B — legacy model, 8k context. Prefer llama_3_1_70b_versatile.

Llama 3 8B — legacy model, 8k context. Prefer llama_3_1_8b_instant.

Llama 3.1 70B — powerful, 128k context.

Llama 3.1 8B — fast, lightweight, 128k context. Good for high-throughput.

Llama 3.2 11B Vision — multimodal model supporting image inputs.

Llama 3.2 1B — ultra-lightweight model for edge use cases.

Llama 3.2 3B — small but capable.

Llama 3.2 90B Vision — large multimodal model.

Llama 3.3 70B with speculative decoding for faster output.

Llama 3.3 70B — latest generation, 128k context. Recommended default.

Mixtral 8x7B MoE — 32k context. Strong at reasoning and code.

Qwen QwQ 32B — strong reasoning and long-context model.