feature about chat completions reasoning, support gemini-3-pro works ok and support claude model enable thinking

src/lib/api-config.ts

@@ -7,16 +7,15 @@ export const standardHeaders = () => ({
   accept: "application/json",
 })
 
-const COPILOT_VERSION = "0.26.7"
+const COPILOT_VERSION = "0.35.0"
 const EDITOR_PLUGIN_VERSION = `copilot-chat/${COPILOT_VERSION}`
 const USER_AGENT = `GitHubCopilotChat/${COPILOT_VERSION}`
 
-const API_VERSION = "2025-04-01"
+const API_VERSION = "2025-10-01"
 
 export const copilotBaseUrl = (state: State) =>
-  state.accountType === "individual" ?
-    "https://api.githubcopilot.com"
-  : `https://api.${state.accountType}.githubcopilot.com`
+  `https://api.${state.accountType}.githubcopilot.com`
+
 export const copilotHeaders = (state: State, vision: boolean = false) => {
   const headers: Record<string, string> = {
     Authorization: `Bearer ${state.copilotToken}`,
@@ -25,7 +24,7 @@ export const copilotHeaders = (state: State, vision: boolean = false) => {
     "editor-version": `vscode/${state.vsCodeVersion}`,
     "editor-plugin-version": EDITOR_PLUGIN_VERSION,
     "user-agent": USER_AGENT,
-    "openai-intent": "conversation-panel",
+    "openai-intent": "conversation-agent",
     "x-github-api-version": API_VERSION,
     "x-request-id": randomUUID(),
     "x-vscode-user-agent-library-version": "electron-fetch",

src/lib/tokenizer.ts

@@ -73,6 +73,9 @@ const calculateMessageTokens = (
   const tokensPerName = 1
   let tokens = tokensPerMessage
   for (const [key, value] of Object.entries(message)) {
+    if (key === "reasoning_opaque") {
+      continue
+    }
     if (typeof value === "string") {
       tokens += encoder.encode(value).length
     }

src/routes/messages/anthropic-types.ts

@@ -56,6 +56,7 @@ export interface AnthropicToolUseBlock {
 export interface AnthropicThinkingBlock {
   type: "thinking"
   thinking: string
+  signature: string
 }
 
 export type AnthropicUserContentBlock =
@@ -196,6 +197,7 @@ export interface AnthropicStreamState {
   messageStartSent: boolean
   contentBlockIndex: number
   contentBlockOpen: boolean
+  thinkingBlockOpen: boolean
   toolCalls: {
     [openAIToolIndex: number]: {
       id: string

src/routes/messages/handler.ts

@@ -60,6 +60,7 @@ export async function handleCompletion(c: Context) {
       contentBlockIndex: 0,
       contentBlockOpen: false,
       toolCalls: {},
+      thinkingBlockOpen: false,
     }
 
     for await (const rawEvent of response) {

src/routes/messages/non-stream-translation.ts

@@ -1,3 +1,6 @@
+import type { Model } from "~/services/copilot/get-models"
+
+import { state } from "~/lib/state"
 import {
   type ChatCompletionResponse,
   type ChatCompletionsPayload,
@@ -29,11 +32,15 @@ import { mapOpenAIStopReasonToAnthropic } from "./utils"
 export function translateToOpenAI(
   payload: AnthropicMessagesPayload,
 ): ChatCompletionsPayload {
+  const modelId = translateModelName(payload.model)
+  const model = state.models?.data.find((m) => m.id === modelId)
+  const thinkingBudget = getThinkingBudget(payload, model)
   return {
-    model: translateModelName(payload.model),
+    model: modelId,
     messages: translateAnthropicMessagesToOpenAI(
       payload.messages,
       payload.system,
+      modelId,
     ),
     max_tokens: payload.max_tokens,
     stop: payload.stop_sequences,
@@ -43,14 +50,36 @@ export function translateToOpenAI(
     user: payload.metadata?.user_id,
     tools: translateAnthropicToolsToOpenAI(payload.tools),
     tool_choice: translateAnthropicToolChoiceToOpenAI(payload.tool_choice),
+    thinking_budget: thinkingBudget,
+  }
+}
+
+function getThinkingBudget(
+  payload: AnthropicMessagesPayload,
+  model: Model | undefined,
+): number | undefined {
+  const thinking = payload.thinking
+  if (model && thinking) {
+    const maxThinkingBudget = Math.min(
+      model.capabilities.supports.max_thinking_budget ?? 0,
+      (model.capabilities.limits.max_output_tokens ?? 0) - 1,
+    )
+    if (maxThinkingBudget > 0 && thinking.budget_tokens !== undefined) {
+      const budgetTokens = Math.min(thinking.budget_tokens, maxThinkingBudget)
+      return Math.max(
+        budgetTokens,
+        model.capabilities.supports.min_thinking_budget ?? 1024,
+      )
+    }
   }
+  return undefined
 }
 
 function translateModelName(model: string): string {
   // Subagent requests use a specific model number which Copilot doesn't support
   if (model.startsWith("claude-sonnet-4-")) {
     return model.replace(/^claude-sonnet-4-.*/, "claude-sonnet-4")
-  } else if (model.startsWith("claude-opus-")) {
+  } else if (model.startsWith("claude-opus-4-")) {
     return model.replace(/^claude-opus-4-.*/, "claude-opus-4")
   }
   return model
@@ -59,13 +88,14 @@ function translateModelName(model: string): string {
 function translateAnthropicMessagesToOpenAI(
   anthropicMessages: Array<AnthropicMessage>,
   system: string | Array<AnthropicTextBlock> | undefined,
+  modelId: string,
 ): Array<Message> {
   const systemMessages = handleSystemPrompt(system)
 
   const otherMessages = anthropicMessages.flatMap((message) =>
     message.role === "user" ?
       handleUserMessage(message)
-    : handleAssistantMessage(message),
+    : handleAssistantMessage(message, modelId),
   )
 
   return [...systemMessages, ...otherMessages]
@@ -125,6 +155,7 @@ function handleUserMessage(message: AnthropicUserMessage): Array<Message> {
 
 function handleAssistantMessage(
   message: AnthropicAssistantMessage,
+  modelId: string,
 ): Array<Message> {
   if (!Array.isArray(message.content)) {
     return [
@@ -139,25 +170,40 @@ function handleAssistantMessage(
     (block): block is AnthropicToolUseBlock => block.type === "tool_use",
   )
 
-  const textBlocks = message.content.filter(
-    (block): block is AnthropicTextBlock => block.type === "text",
-  )
-
-  const thinkingBlocks = message.content.filter(
+  let thinkingBlocks = message.content.filter(
     (block): block is AnthropicThinkingBlock => block.type === "thinking",
   )
 
-  // Combine text and thinking blocks, as OpenAI doesn't have separate thinking blocks
-  const allTextContent = [
-    ...textBlocks.map((b) => b.text),
-    ...thinkingBlocks.map((b) => b.thinking),
-  ].join("\n\n")
+  if (modelId.startsWith("claude")) {
+    thinkingBlocks = thinkingBlocks.filter(
+      (b) =>
+        b.thinking
+        && b.thinking.length > 0
+        && b.signature
+        && b.signature.length > 0
+        // gpt signature has @ in it, so filter those out for claude models
+        && !b.signature.includes("@"),
+    )
+  }
+
+  const thinkingContents = thinkingBlocks
+    .filter((b) => b.thinking && b.thinking.length > 0)
+    .map((b) => b.thinking)
+
+  const allThinkingContent =
+    thinkingContents.length > 0 ? thinkingContents.join("\n\n") : undefined
+
+  const signature = thinkingBlocks.find(
+    (b) => b.signature && b.signature.length > 0,
+  )?.signature
 
   return toolUseBlocks.length > 0 ?
       [
         {
           role: "assistant",
-          content: allTextContent || null,
+          content: mapContent(message.content),
+          reasoning_text: allThinkingContent,
+          reasoning_opaque: signature,
           tool_calls: toolUseBlocks.map((toolUse) => ({
             id: toolUse.id,
             type: "function",
@@ -172,6 +218,8 @@ function handleAssistantMessage(
         {
           role: "assistant",
           content: mapContent(message.content),
+          reasoning_text: allThinkingContent,
+          reasoning_opaque: signature,
         },
       ]
 }
@@ -191,11 +239,8 @@ function mapContent(
   const hasImage = content.some((block) => block.type === "image")
   if (!hasImage) {
     return content
-      .filter(
-        (block): block is AnthropicTextBlock | AnthropicThinkingBlock =>
-          block.type === "text" || block.type === "thinking",
-      )
-      .map((block) => (block.type === "text" ? block.text : block.thinking))
+      .filter((block): block is AnthropicTextBlock => block.type === "text")
+      .map((block) => block.text)
       .join("\n\n")
   }
 
@@ -204,12 +249,6 @@ function mapContent(
     switch (block.type) {
       case "text": {
         contentParts.push({ type: "text", text: block.text })
-
-        break
-      }
-      case "thinking": {
-        contentParts.push({ type: "text", text: block.thinking })
-
         break
       }
       case "image": {
@@ -219,7 +258,6 @@ function mapContent(
             url: `data:${block.source.media_type};base64,${block.source.data}`,
           },
         })
-
         break
       }
       // No default
@@ -282,34 +320,32 @@ export function translateToAnthropic(
   response: ChatCompletionResponse,
 ): AnthropicResponse {
   // Merge content from all choices
-  const allTextBlocks: Array<AnthropicTextBlock> = []
-  const allToolUseBlocks: Array<AnthropicToolUseBlock> = []
-  let stopReason: "stop" | "length" | "tool_calls" | "content_filter" | null =
-    null // default
-  stopReason = response.choices[0]?.finish_reason ?? stopReason
+  const assistantContentBlocks: Array<AnthropicAssistantContentBlock> = []
+  let stopReason = response.choices[0]?.finish_reason ?? null
 
   // Process all choices to extract text and tool use blocks
   for (const choice of response.choices) {
     const textBlocks = getAnthropicTextBlocks(choice.message.content)
+    const thinkBlocks = getAnthropicThinkBlocks(
+      choice.message.reasoning_text,
+      choice.message.reasoning_opaque,
+    )
     const toolUseBlocks = getAnthropicToolUseBlocks(choice.message.tool_calls)
 
-    allTextBlocks.push(...textBlocks)
-    allToolUseBlocks.push(...toolUseBlocks)
+    assistantContentBlocks.push(...thinkBlocks, ...textBlocks, ...toolUseBlocks)
 
     // Use the finish_reason from the first choice, or prioritize tool_calls
     if (choice.finish_reason === "tool_calls" || stopReason === "stop") {
       stopReason = choice.finish_reason
     }
   }
 
-  // Note: GitHub Copilot doesn't generate thinking blocks, so we don't include them in responses
-
   return {
     id: response.id,
     type: "message",
     role: "assistant",
     model: response.model,
-    content: [...allTextBlocks, ...allToolUseBlocks],
+    content: assistantContentBlocks,
     stop_reason: mapOpenAIStopReasonToAnthropic(stopReason),
     stop_sequence: null,
     usage: {
@@ -329,7 +365,7 @@ export function translateToAnthropic(
 function getAnthropicTextBlocks(
   messageContent: Message["content"],
 ): Array<AnthropicTextBlock> {
-  if (typeof messageContent === "string") {
+  if (typeof messageContent === "string" && messageContent.length > 0) {
     return [{ type: "text", text: messageContent }]
   }
 
@@ -342,6 +378,31 @@ function getAnthropicTextBlocks(
   return []
 }
 
+function getAnthropicThinkBlocks(
+  reasoningText: string | null | undefined,
+  reasoningOpaque: string | null | undefined,
+): Array<AnthropicThinkingBlock> {
+  if (reasoningText && reasoningText.length > 0) {
+    return [
+      {
+        type: "thinking",
+        thinking: reasoningText,
+        signature: reasoningOpaque || "",
+      },
+    ]
+  }
+  if (reasoningOpaque && reasoningOpaque.length > 0) {
+    return [
+      {
+        type: "thinking",
+        thinking: "",
+        signature: reasoningOpaque,
+      },
+    ]
+  }
+  return []
+}
+
 function getAnthropicToolUseBlocks(
   toolCalls: Array<ToolCall> | undefined,
 ): Array<AnthropicToolUseBlock> {