Spring AI 2.0.0-M5: Embedding fails with OpenAI-compatible APIs (Base64 string vs float[])

[arianox]

Bug description

Spring AI 2.0.0-M5 fails when using OpenAI-compatible embedding APIs (e.g. IONOS bge-m3) due to a mismatch in embedding response format.

The error indicates that the embedding is returned as a string (likely Base64) instead of a float[], which causes deserialization to fail:
json: cannot unmarshal string into Go struct field Embedding.data.embedding of type []float32

This worked correctly in 2.0.0-M4.

---

Environment

• Spring AI: 2.0.0-M5
• Java: 21
• Spring Boot: 4.0.6
• Vector Store: Qdrant
• Dependencies:
  • spring-ai-starter-model-openai
  • spring-ai-qdrant-store
  • spring-ai-rag
• Provider: IONOS OpenAI-compatible endpoint
• Model: bge-m3

---

Steps to reproduce

1. Configure Spring AI with an OpenAI-compatible endpoint (e.g. IONOS)
2. Use spring-ai-starter-model-openai
3. Call embeddings

[sdeleuze]

Closing in favor of the related PR #5902.