Added the exact token count where the provider reports it

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionAPIRequest.cs

@@ -23,6 +23,17 @@ public record ChatCompletionAPIRequest(
 
     [JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
     public bool? ParallelToolCalls { get; init; }
+
+    /// <summary>
+    /// Asks a streamed request to end with what it cost.
+    /// </summary>
+    /// <remarks>
+    /// Derived rather than set, so that every provider which builds one of these asks for it
+    /// without having to know that it exists. A request which is not streamed carries no such
+    /// line, and then the block would only be a field the provider has to ignore.
+    /// </remarks>
+    [JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+    public ChatCompletionStreamOptions? StreamOptions => this.Stream ? ChatCompletionStreamOptions.INCLUDE_USAGE : null;
     
     // Attention: The "required" modifier is not supported for [JsonExtensionData].
     [JsonExtensionData]

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionDeltaStreamLine.cs

@@ -15,12 +15,30 @@ public record ChatCompletionDeltaStreamLine(string Id, string Object, uint Creat
     {
     }
     
+    /// <summary>
+    /// What the provider says the request cost, on the one line which carries it.
+    /// </summary>
+    /// <remarks>
+    /// Not a positional parameter: every provider builds an empty line through the constructor
+    /// above, and a further parameter would change all of those call sites for a value none of
+    /// them has. Providers send this block only when the request asked for it, and then on a final
+    /// line of its own which carries no choices -- which is why the usage is read apart from the
+    /// content rather than next to it.
+    /// </remarks>
+    public ChatCompletionUsage? Usage { get; init; }
+
     /// <inheritdoc />
     public bool ContainsContent() => this.Choices.Count > 0;
 
     /// <inheritdoc />
     public ContentStreamChunk GetContent() => new(this.Choices[0].Delta.Content, []);
 
+    /// <inheritdoc />
+    public bool ContainsUsage() => this.GetUsage().IsKnown;
+
+    /// <inheritdoc />
+    public TokenUsage GetUsage() => this.Usage is null ? TokenUsage.UNKNOWN : TokenUsage.OfReported(this.Usage.PromptTokens, this.Usage.CompletionTokens);
+
     #region Implementation of IAnnotationStreamLine
 
     //

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionStreamOptions.cs

@@ -0,0 +1,18 @@
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// What a streamed chat completion should report beyond its content.
+/// </summary>
+/// <remarks>
+/// An OpenAI-compatible provider says nothing about what a streamed request cost unless it is asked
+/// to. Without this block, the stream simply ends and the only token number anybody ever sees is
+/// the one AI Studio estimated for itself.
+/// </remarks>
+/// <param name="IncludeUsage">Whether the stream should end with a line stating the token usage.</param>
+public sealed record ChatCompletionStreamOptions(bool IncludeUsage)
+{
+    /// <summary>
+    /// Asks for the usage line.
+    /// </summary>
+    public static readonly ChatCompletionStreamOptions INCLUDE_USAGE = new(true);
+}

app/MindWork AI Studio/Provider/OpenAI/ChatCompletionUsage.cs

@@ -0,0 +1,30 @@
+// ReSharper disable ClassNeverInstantiated.Global
+namespace AIStudio.Provider.OpenAI;
+
+/// <summary>
+/// What an OpenAI-compatible provider reports a chat completion cost.
+/// </summary>
+/// <remarks>
+/// Every number is optional because this is somebody else's JSON: the block arrives only when the
+/// request asked for it, and the providers which follow the shape loosely leave fields out. Reading
+/// it is one thing, believing it another -- <see cref="TokenUsage.OfReported"/> decides that.
+/// </remarks>
+public sealed record ChatCompletionUsage
+{
+    /// <summary>
+    /// What everything sent to the model cost.
+    /// </summary>
+    public int? PromptTokens { get; init; }
+
+    /// <summary>
+    /// What the model wrote in answer.
+    /// </summary>
+    public int? CompletionTokens { get; init; }
+
+    /// <summary>
+    /// What the provider says both of them add up to. Read but not relied upon: it is the sum of
+    /// the other two wherever a provider fills all three, and this way a provider which sends only
+    /// this one is not a reason to throw the other numbers away.
+    /// </summary>
+    public int? TotalTokens { get; init; }
+}