Does finishReason "stop" always send for streamText?

does finishReason "stop" get sent always in the case of a successful generation? for streamText, i mean.

Now I have a complete picture. Let me summarize the findings:

##Summary: finishReason in streamText

No, finishReason: "stop" is NOT always sent for a successful generation. Here's how it works:

###Possible finishReason values:

From packages/ai/src/types/language-model.ts:

export type FinishReason =
  | 'stop'        // model stopped naturally (e.g., end of response)
  | 'length'      // model stopped due to max tokens
  | 'content-filter' // model stopped due to content filtering
  | 'tool-calls'  // model stopped to make tool calls
  | 'error'       // model stopped due to an error
  | 'other';      // model stopped for other reasons

###When you get each finishReason:

1. "stop" - The model naturally completed its generation (most common for a simple successful text completion)

2. "tool-calls" - The model stopped because it wants to call tools. This is a "successful" generation too, just one that requires tool execution. If you have multi-step tool loops, intermediate steps will have finishReason: "tool-calls" and only the final step will typically have "stop".

3. "length" - The model hit the maximum token limit (maxTokens)

4. "content-filter" - The provider's content filter triggered

5. "error" - An error occurred during generation

6. "other" - Default fallback when the provider doesn't specify a reason (see packages/openai/src/chat/map-openai-finish-reason.ts)

###Default behavior:

• Providers initialize finishReason as { unified: 'other', raw: undefined } (line 443 in the OpenAI provider)
• If no finish chunk comes from the provider with a reason, or the stream ends early, you may get 'other'
• In stream-text.ts (line 974): const finishReason = recordedFinishReason ?? 'other'; — if no finish reason was recorded, it defaults to 'other'

###Key takeaway:

A "successful generation" with tool calling will often end with finishReason: "tool-calls" on intermediate steps. The "stop" reason indicates the model naturally finished generating text without needing to call tools or hitting any limits.