Phi 3.5 MoE

Phi-3-Vision is Microsoft’s compact, open-weight multimodal model that understands images + text and answers in text. Optimized for documents, charts, UI screenshots, diagrams, and photos, it delivers strong OCR and visual reasoning in a small footprint suitable for single-GPU or edge deployment.

Phi-4-mini-reasoning is a lightweight open model built upon synthetic data with a focus on high-quality, reasoning dense data further finetuned for more advanced math reasoning capabilities. The model belongs to the Phi-4 model family and supports 128K token context length.

Phi-4-reasoning-plus is an open-weight reasoning model derived from Phi-4, trained with chain-of-thought supervised fine-tuning and extra reinforcement learning on high-quality synthetic and filtered public-domain data (math, science, coding) with safety alignment. It delivers stronger reasoning in a small model, at the cost of ~50% longer outputs and higher latency.

A 32B-parameter GLM-4 checkpoint with 128k context. Strong general reasoning and coding, tool and function calling, and clean JSON for long-form tasks.

Anthropic’s Sonnet 4.6 model for scaled production and complex tasks across coding, agents, and professional workflows.

Nova Premier is the flagship reasoning model in the series. It delivers the highest accuracy for analysis, coding, and math, with very long context, robust tool calling, and strict JSON guarantees for high stakes workflows.