Azure AI Foundry — DeepSeek-V4-Flash: is an Anthropic-compatible endpoint (/v1/messages) available for my deployment?

I have a DeepSeek-V4-Flash model deployed on Azure AI Foundry (France Central), endpoint https://apideep.services.ai.azure.com.

The OpenAI-compatible route works fine: POST /openai/v1/chat/completions returns 200 (chat + tool-calling OK).

I need to call this model through an Anthropic-compatible API (/v1/messages format), because the agent framework I use (opencode) only handles DeepSeek reasoning models reliably through that format — through the OpenAI-compatible route, the reasoning content is dropped between turns and the agent loops on multi-turn tool calls.

DeepSeek's public API exposes this (https://api.deepseek.com/anthropic/v1/messages), but I must stay on Azure (EU) for GDPR compliance.

Questions:

1. Does an Azure AI Foundry DeepSeek deployment expose (or can it expose) an Anthropic-compatible endpoint (/anthropic/v1/messages)? It currently returns 404.
2. If not, is there a supported, stable way to disable the model's reasoning output (e.g. reasoning_effort)?
3. Do you offer an instruct / non-reasoning DeepSeek variant in an EU region?