Expected Behavior Currently, many model providers offer reason models, and users choose whether to use them depending on the specific use case. Spring AI needs to improve its support for reason models. add reasoningContent to AssistantMessage
public class AssistantMessage extends AbstractMessage implements MediaContent {
private final List<ToolCall> toolCalls;
protected final List<Media> media;
private String reasoningContent
// other codes
...
Current Behavior At present, supporting reason models requires adapting to different model vendors. This makes it difficult to standardize and unify reasoning messages, and also results in relatively high adaptation costs.
Comment From: simsnow
+1 The openai chatModel support toolCall and media,but not support reasoningContent, The deepseek chatModel support reasoningContent,but not support toolCall and media
Comment From: sha2kyou
+1 The openai chatModel support toolCall and media,but not support reasoningContent, The deepseek chatModel support reasoningContent,but not support toolCall and media
Same with Bedrock converse models—it doesn’t support reasoning either, but support tool calling.
Comment From: AllyBully
+1 Now I'm recording the reasoning content into the database, and I still need to rewrite a lot of content to adapt to the reasoningContent of different large models.