Model catalog

Claude Haiku 4.5 is Anthropic's fast, cost-efficient Claude model for high-volume chat and tool-using workloads via AWS Bedrock.

Claude Sonnet 4.6 is Anthropic's balanced Claude model for coding, reasoning, and agentic workloads via AWS Bedrock Global CRIS.

DeepSeek V3.2 is a large MoE chat model tuned for strong reasoning and coding at competitive cost, served via Azure AI Foundry.

Phi-4 is a compact open-weight model from Microsoft Research, efficient for text tasks. Note: latency can be high under load on shared Azure Foundry capacity.

Phi-4 Mini Instruct is a lightweight open-weight Microsoft model optimized for efficient instruction following and chat on Azure AI Foundry.

GPT-4.1 is OpenAI's flagship multimodal model with a very large context window, strong instruction following, and vision support.

GPT-4.1 Mini is a lower-cost GPT-4.1-series chat model with a large context window, strong instruction following, and structured-output support via Azure AI Foundry.

GPT-5.4 is OpenAI's latest flagship chat model with an extended context window and multimodal capabilities via Azure OpenAI.

GPT-5.4 Nano is OpenAI's lightweight GPT-5.4 variant for low-latency, cost-efficient classification, extraction, and high-volume chat via Azure OpenAI.

GPT-5.5 is OpenAI's newer flagship chat model with extended context, reasoning, and multimodal capabilities via Azure OpenAI.

GPT-5.6 Luna is the fastest and lowest-cost GPT-5.6 variant for high-volume and cost-sensitive routing via Azure OpenAI.

GPT-5.6 Sol is OpenAI's flagship GPT-5.6 variant for complex professional reasoning, coding, and agentic workloads via Azure OpenAI.

GPT-5.6 Terra balances GPT-5.6 intelligence, speed, and cost for general production workloads via Azure OpenAI.

o3 is OpenAI's reasoning model with multimodal input. Uses ``max_completion_tokens`` instead of ``max_tokens`` on the chat API.

Grok 4 is xAI's frontier model with multimodal input, long context, and strong general reasoning served through Azure AI Foundry.

Grok 4.1 Fast Reasoning is a latency-optimized variant of Grok 4 prioritizing speed while retaining strong chain-of-thought performance.