pi-provider-qoder-cn 0.2.28 → 0.2.30

package/README.md

@@ -193,9 +193,9 @@ The provider list is re-registered in the running session, so the change is usab
 in `/model` immediately — no restart. Append `list` to print the models pi currently
 has cached, without touching the network.
 
-Models offered only in the `app` group are included too. QoDuck (`lmodel`), for
-example, is a selectable free model in both regions that the `chat` group omits;
-entries already present in `chat` are not duplicated, and UI-only tiers from other
+Models offered only in the `app` group are included too �� QoDuck (`lmodel`) was
+such an entry before its upstream removal in the 2026-11 catalog. Entries
+already present in `chat` are not duplicated, and UI-only tiers from other
 groups (`Performance`, `Premium`, ...) stay out of the picker.
 
 The catalog key is a slot, not a model: upstream has served Kimi-K2.6 through
@@ -284,22 +284,46 @@ Choose the Qoder CN endpoint (default: personal account)
 
 ## Models
 
+### Qoder CN (`qoder-cn`) �� catalog as of 2026-11-11
+
+| Model | id | Context | Thinking effort |
+|---|---|---|---|
+| Auto | `auto` | 1M | �� |
+| Qwen 3.8-Max | `qwen3.8-max` | 1M | ? |
+| Qwen 3.8-Flash | `qwen3.8-flash` | 1M | ? |
+| DeepSeek V4 Pro | `deepseek-v4-pro` | 1M | ? |
+| DeepSeek Flash | `deepseek-flash` | 1M | ? |
+| GLM-5.3 | `glm-5.3` | 1M | ? |
+| GLM-5.3-Flash | `glm-5.3-flash` | 1M | ? |
+| Kimi-K3.1-Pro | `kimi-k3.1-pro` | 1M | ? |
+| Kimi-K2.8-Preview | `kimi-k2.8-preview` | 1M | ? |
+| Step-5-Preview | `step-5-preview` | 1M | ? |
+| MiniMax-M3.1-Flash | `minimax-m3.1-flash` | 1M | ? |
+
+### Qoder Global (`qoder`) �� catalog as of 2026-11-11
+
 | Model | id | Context | Thinking effort |
 |---|---|---|---|
-| Auto | `auto` | 200K | — |
-| Qwen 3.8-Max | `qwen3.8-max` | 1M | ✅ |
-| Qwen 3.8-Flash | `qwen3.8-flash` | 1M | ✅ |
-| Qwen 3.7-Max | `qwen3.7-max` | 1M | — |
-| Qwen 3.7-Plus | `qwen3.7-plus` | 1M | — |
-| Qwen 3.7-Flash | `qwen3.7-flash` | 1M | — |
-| DeepSeek V4 Pro | `deepseek-v4-pro` | 1M | ✅ |
-| DeepSeek V4 Flash | `deepseek-v4-flash` | 1M | ✅ |
-| GLM-5.3 | `glm-5.3` | 1M | ✅ |
-| GLM-5.3-Flash | `glm-5.3-flash` | 1M | ✅ |
-| GLM 5.2 | `glm-5.2` | 1M | ✅ |
-| Kimi-K3 | `kimi-k3` | 1M | ✅ |
-| Kimi-K2.7-Code | `kimi-k2.7-code` | 256K | — |
-| MiniMax M2.7 | `minimax-m2.7` | 200K | — |
+| Auto | `Auto` | 1M | �� |
+| Ultimate | `Ultimate` | 1M | ? |
+| Efficient | `Efficient` | 200K | �� |
+| Sonus | `Sonus` | 1M | ? |
+| Cantus | `Cantus` | 1M | ? |
+| Qwen4-Max-Preview | `Qwen4-Max-Preview` | 1M | ? |
+| Qwen 3.8-Max | `Qwen3.8-Max` | 1M | ? |
+| Qwen 3.8-Flash | `Qwen3.8-Flash` | 1M | ? |
+| Kimi-K3.1-Pro | `Kimi-K3.1-Pro` | 1M | ? |
+| Kimi-K2.8-Preview | `Kimi-K2.8-Preview` | 1M | ? |
+| GLM-5.3 | `GLM-5.3` | 1M | ? |
+| GLM-5.3-Flash | `GLM-5.3-Flash` | 1M | ? |
+| DeepSeek-V4-Pro | `DeepSeek-V4-Pro` | 1M | ? |
+| DeepSeek-Flash | `DeepSeek-Flash` | 1M | ? |
+| MiniMax-M3.1-Flash | `MiniMax-M3.1-Flash` | 1M | ? |
+
+Retired in the 2026-11 catalog refresh: QoDuck (`qoduck`/`QoDuck`, both
+regions) and the Qwen3.7 / GLM-5.2 / Kimi-K3 / Kimi-K2.7-Code / MiniMax-M3 /
+DeepSeek-V4-Flash families. Run `/qoder-cn.models` or `/qoder.models` for the
+live list �� the tables above are the offline snapshot, not the source of truth.
 
 ```bash
 pi --provider qoder-cn --model qwen3.8-flash

package/dist/index.js

@@ -696,7 +696,7 @@ var init_catalog = __esm({
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: true,
+        reasoning: false,
         supportsEffort: false,
         input: ["text", "image"],
         cost: ZERO_COST,
@@ -718,65 +718,65 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Performance",
-        upstreamKey: "performance",
-        name: "Performance",
+        id: "Efficient",
+        upstreamKey: "efficient",
+        name: "Efficient",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: true,
-        supportsEffort: true,
+        reasoning: false,
+        supportsEffort: false,
         input: ["text", "image"],
         cost: ZERO_COST,
-        contextWindow: DEFAULT_CONTEXT_WINDOW,
+        contextWindow: 2e5,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Efficient",
-        upstreamKey: "efficient",
-        name: "Efficient",
+        id: "Sonus",
+        upstreamKey: "smodel",
+        name: "Sonus",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Lite",
-        upstreamKey: "lite",
-        name: "Lite",
+        id: "Cantus",
+        upstreamKey: "cmodel",
+        name: "Cantus",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
-        input: ["text"],
+        reasoning: true,
+        supportsEffort: true,
+        input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Qwen3.7Plus",
-        upstreamKey: "qmodel",
-        name: "Qwen3.7 Plus",
+        id: "Qwen4-Max-Preview",
+        upstreamKey: "qmodel_latest",
+        name: "Qwen4-Max-Preview",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Cantus",
-        upstreamKey: "cmodel",
-        name: "Cantus",
+        id: "Qwen3.8-Max",
+        upstreamKey: "qmodel_38max",
+        name: "Qwen3.8-Max",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
@@ -788,9 +788,9 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Qwen3.8-Max",
-        upstreamKey: "qmodel_preview",
-        name: "Qwen3.8-Max",
+        id: "Qwen3.8-Flash",
+        upstreamKey: "qfmodel",
+        name: "Qwen3.8-Flash",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
@@ -802,23 +802,23 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Qwen3.7-Max",
-        upstreamKey: "qmodel_latest",
-        name: "Qwen3.7-Max",
+        id: "Kimi-K3.1-Pro",
+        upstreamKey: "kmodel_latest",
+        name: "Kimi-K3.1-Pro",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "DeepSeek-V4-Pro",
-        upstreamKey: "dmodel",
-        name: "DeepSeek-V4-Pro",
+        id: "Kimi-K2.8-Preview",
+        upstreamKey: "kmodel",
+        name: "Kimi-K2.8-Preview",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
@@ -830,9 +830,9 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "DeepSeek-V4-Flash",
-        upstreamKey: "dfmodel",
-        name: "DeepSeek-V4-Flash",
+        id: "GLM-5.3",
+        upstreamKey: "gmodel",
+        name: "GLM-5.3",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
@@ -844,9 +844,9 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "GLM-5.2",
-        upstreamKey: "gm51model",
-        name: "GLM-5.2",
+        id: "GLM-5.3-Flash",
+        upstreamKey: "gfmodel",
+        name: "GLM-5.3-Flash",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
@@ -858,43 +858,42 @@ var init_catalog = __esm({
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Kimi-K2.7-Code",
-        upstreamKey: "kmodel",
-        name: "Kimi-K2.7-Code",
+        id: "DeepSeek-V4-Pro",
+        upstreamKey: "dmodel",
+        name: "DeepSeek-V4-Pro",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
-        // Catalog advertises 256K; not included in the 1M live test in issue #13.
-        contextWindow: 256e3,
+        contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "Kimi-K3",
-        upstreamKey: "kmodel_latest",
-        name: "Kimi-K3",
+        id: "DeepSeek-Flash",
+        upstreamKey: "dfmodel",
+        name: "DeepSeek-Flash",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS
       },
       {
-        id: "MiniMax-M3",
+        id: "MiniMax-M3.1-Flash",
         upstreamKey: "mmodel",
-        name: "MiniMax-M3",
+        name: "MiniMax-M3.1-Flash",
         api: "qoder-api",
         provider: "qoder",
         baseUrl: getQoderBaseUrl("global"),
-        reasoning: false,
-        supportsEffort: false,
+        reasoning: true,
+        supportsEffort: true,
         input: ["text", "image"],
         cost: ZERO_COST,
         contextWindow: DEFAULT_CONTEXT_WINDOW,
@@ -913,10 +912,9 @@ var init_catalog = __esm({
         supportsEffort: false,
         input: ["text", "image"],
         cost: ZERO_COST,
-        // CN Auto has not been live-tested at 1M; the live catalog reports 200K.
-        contextWindow: 2e5,
+        contextWindow: DEFAULT_CONTEXT_WINDOW,
         maxTokens: MAX_OUTPUT_TOKENS,
-        description: "Qoder CN smart routing; live catalog reports 200K max input."
+        description: "Qoder CN smart routing; the 2026-11 catalog advertises 1M."
       }
     ];
     PI_THINKING_LEVELS = ["minimal", "low", "medium", "high", "xhigh", "max"];
@@ -3008,6 +3006,21 @@ var THINKING_TAG_VARIANTS = [
   { open: "<thought>", close: "</thought>" }
 ];
 var ALL_THINKING_TAGS = THINKING_TAG_VARIANTS.flatMap((variant) => [variant.open, variant.close]);
+function hasAnyCompleteThinkingTag(text) {
+  return THINKING_TAG_VARIANTS.some((variant) => text.includes(variant.open) || text.includes(variant.close));
+}
+function hasOrphanThinkingCloser(text) {
+  for (const variant of THINKING_TAG_VARIANTS) {
+    const closePos = text.indexOf(variant.close);
+    if (closePos === -1) continue;
+    const openerBefore = THINKING_TAG_VARIANTS.some((other) => {
+      const openPos = text.indexOf(other.open);
+      return openPos !== -1 && openPos < closePos;
+    });
+    if (!openerBefore) return true;
+  }
+  return false;
+}
 function getTrailingPossibleTagPrefixLength(text, tag) {
   const maxPrefixLength = Math.min(text.length, tag.length - 1);
   for (let len = maxPrefixLength; len > 0; len--) {
@@ -3055,6 +3068,23 @@ var ThinkingTagParser = class {
   textBlockIndex = null;
   lastTextBlockIndex = null;
   activeEndTag = THINKING_TAG_VARIANTS[0].close;
+  orphanCloserMeansThinking = false;
+  /**
+   * Must be called before the first `processChunk`.
+   *
+   * When the reasoning_content channel carried no thinking body this turn
+   * (only a bare opener tag, or nothing at all), an orphan closer in the
+   * content stream is preceded by THINKING text rather than answer text.
+   * stream.ts decides this once, at the first content delta, because Qoder
+   * sends the reasoning channel before the content channel.
+   *
+   * Default false preserves the historical behaviour: the opener and its
+   * thinking text arrived via reasoning_content, so text before the orphan
+   * closer in the content stream is visible answer text.
+   */
+  setOrphanCloserMeansThinking(value) {
+    this.orphanCloserMeansThinking = value;
+  }
   processChunk(chunk) {
     this.textBuffer += chunk;
     while (this.textBuffer.length > 0) {
@@ -3091,6 +3121,17 @@ var ThinkingTagParser = class {
         content: block.thinking,
         partial: this.output
       });
+    } else if (this.orphanCloserMeansThinking && !this.thinkingExtracted) {
+      this.emitThinking(this.textBuffer);
+      if (this.thinkingBlockIndex !== null) {
+        const block = this.output.content[this.thinkingBlockIndex];
+        this.stream.push({
+          type: "thinking_end",
+          contentIndex: this.thinkingBlockIndex,
+          content: block.thinking,
+          partial: this.output
+        });
+      }
     } else {
       this.emitText(this.textBuffer);
     }
@@ -3124,6 +3165,27 @@ var ThinkingTagParser = class {
       return;
     }
     if (bestCloseVariant !== null) {
+      if (this.orphanCloserMeansThinking && bestClosePos > 0) {
+        const thinkingText = this.textBuffer.slice(0, bestClosePos);
+        this.emitThinking(thinkingText);
+        if (this.thinkingBlockIndex !== null) {
+          const block = this.output.content[this.thinkingBlockIndex];
+          this.stream.push({
+            type: "thinking_end",
+            contentIndex: this.thinkingBlockIndex,
+            content: block.thinking,
+            partial: this.output
+          });
+          this.thinkingBlockIndex = null;
+        }
+        this.textBuffer = this.textBuffer.slice(bestClosePos + bestCloseVariant.close.length);
+        if (this.textBuffer.startsWith("\n\n")) this.textBuffer = this.textBuffer.slice(2);
+        else if (this.textBuffer.startsWith("\n")) this.textBuffer = this.textBuffer.slice(1);
+        this.thinkingExtracted = true;
+        this.lastTextBlockIndex = this.textBlockIndex;
+        this.textBlockIndex = null;
+        return;
+      }
       if (bestClosePos > 0) this.emitText(this.textBuffer.slice(0, bestClosePos));
       this.textBuffer = this.textBuffer.slice(bestClosePos + bestCloseVariant.close.length);
       if (this.textBuffer.startsWith("\n\n")) this.textBuffer = this.textBuffer.slice(2);
@@ -3133,7 +3195,9 @@ var ThinkingTagParser = class {
     const trailingPrefixLength = getMaxTrailingPossibleTagPrefixLength(this.textBuffer, ALL_THINKING_TAGS);
     const safeLen = this.textBuffer.length - trailingPrefixLength;
     if (safeLen > 0) {
-      this.emitText(this.textBuffer.slice(0, safeLen));
+      const safeText = this.textBuffer.slice(0, safeLen);
+      if (this.orphanCloserMeansThinking) this.emitThinking(safeText);
+      else this.emitText(safeText);
       this.textBuffer = this.textBuffer.slice(safeLen);
     }
   }
@@ -3357,6 +3421,7 @@ function transformMessagesForQoder(messages) {
 // src/protocol/stream.ts
 var healCooldowns = /* @__PURE__ */ new Map();
 var HEAL_COOLDOWN_MS = 6e4;
+var CONTENT_CLASSIFY_BUFFER_LIMIT = 8 * 1024;
 function stableHash(prefix, ...inputs) {
   const hash = crypto3.createHash("sha256");
   hash.update(prefix);
@@ -3739,13 +3804,43 @@ function streamQoder(model, context, options) {
         const toolCallsState = [];
         const thinkingEnabled = options?.reasoning !== false && options?.reasoning !== "off";
         const thinkingParser = thinkingEnabled ? new ThinkingTagParser(output, stream) : null;
+        let sawDone = false;
+        let sawFinishReason = false;
+        let sawAnyDelta = false;
+        let reasoningChannelHadBody = false;
+        let contentClassificationDone = false;
+        let contentBuffer = "";
+        const flushContentBuffer = (final = false) => {
+          if (!contentBuffer) return;
+          if (!contentClassificationDone && (final || hasAnyCompleteThinkingTag(contentBuffer) || contentBuffer.length >= CONTENT_CLASSIFY_BUFFER_LIMIT)) {
+            if (thinkingParser && !reasoningChannelHadBody && hasOrphanThinkingCloser(contentBuffer)) {
+              thinkingParser.setOrphanCloserMeansThinking(true);
+            }
+            contentClassificationDone = true;
+          }
+          if (!contentClassificationDone) return;
+          const pending = contentBuffer;
+          contentBuffer = "";
+          if (thinkingParser) thinkingParser.processChunk(pending);
+          else emitPlainText(pending);
+        };
+        const emitPlainText = (text) => {
+          if (!text) return;
+          if (contentBlockIndex === -1) {
+            contentBlockIndex = output.content.length;
+            output.content.push({ type: "text", text: "" });
+            stream.push({ type: "text_start", contentIndex: contentBlockIndex, partial: output });
+          }
+          const block = output.content[contentBlockIndex];
+          block.text += text;
+          stream.push({ type: "text_delta", contentIndex: contentBlockIndex, delta: text, partial: output });
+        };
         if (!started) {
           started = true;
         } else {
           output.content.length = 0;
         }
         stream.push({ type: "start", partial: output });
-        let sawDone = false;
         while (!sawDone) {
           const { done, value } = await reader.read();
           if (done) break;
@@ -3796,9 +3891,11 @@ function streamQoder(model, context, options) {
                 const choice = inner.choices[0];
                 const delta = choice.delta;
                 if (delta) {
+                  sawAnyDelta = true;
                   if (delta.reasoning_content) {
                     rawReasoningTail = (rawReasoningTail + delta.reasoning_content).slice(-512);
                     const reasoningChunk = stripDsmlResidue(stripThinkingTags(delta.reasoning_content));
+                    if (reasoningChunk.trim().length > 0) reasoningChannelHadBody = true;
                     if (reasoningChunk) {
                       if (thinkingBlockIndex === -1) {
                         thinkingBlockIndex = output.content.length;
@@ -3827,24 +3924,14 @@ function streamQoder(model, context, options) {
                       thinkingBlockIndex = -1;
                     }
                     if (thinkingParser) {
-                      thinkingParser.processChunk(delta.content);
+                      contentBuffer += delta.content;
+                      flushContentBuffer();
                     } else {
-                      if (contentBlockIndex === -1) {
-                        contentBlockIndex = output.content.length;
-                        output.content.push({ type: "text", text: "" });
-                        stream.push({ type: "text_start", contentIndex: contentBlockIndex, partial: output });
-                      }
-                      const block = output.content[contentBlockIndex];
-                      block.text += delta.content;
-                      stream.push({
-                        type: "text_delta",
-                        contentIndex: contentBlockIndex,
-                        delta: delta.content,
-                        partial: output
-                      });
+                      emitPlainText(delta.content);
                     }
                   }
                   if (delta.tool_calls && Array.isArray(delta.tool_calls)) {
+                    flushContentBuffer(true);
                     for (const tc of delta.tool_calls) {
                       const idx = tc.index ?? 0;
                       if (!toolCallsState[idx]) {
@@ -3883,6 +3970,7 @@ function streamQoder(model, context, options) {
                   }
                 }
                 if (choice.finish_reason) {
+                  sawFinishReason = true;
                   output.stopReason = choice.finish_reason;
                 }
               }
@@ -3899,6 +3987,7 @@ function streamQoder(model, context, options) {
         }
         await reader.cancel().catch(() => {
         });
+        flushContentBuffer(true);
         if (thinkingParser) {
           thinkingParser.finalize();
         }
@@ -3944,6 +4033,32 @@ function streamQoder(model, context, options) {
           stream.end();
           return;
         }
+        if (!sawAnyDelta) {
+          if (!options?.signal?.aborted) {
+            output.stopReason = "error";
+            output.errorMessage = "Qoder server error: incomplete stream (no content deltas received)";
+            stream.push({ type: "error", reason: "error", error: output });
+            stream.end();
+            return;
+          }
+          output.stopReason = "aborted";
+          stream.push({ type: "error", reason: "aborted", error: output });
+          stream.end();
+          return;
+        }
+        if (!sawDone && !sawFinishReason) {
+          if (!options?.signal?.aborted) {
+            output.stopReason = "error";
+            output.errorMessage = "Qoder server error: incomplete stream (response ended without [DONE] or finish_reason)";
+            stream.push({ type: "error", reason: "error", error: output });
+            stream.end();
+            return;
+          }
+          output.stopReason = "aborted";
+          stream.push({ type: "error", reason: "aborted", error: output });
+          stream.end();
+          return;
+        }
         stream.push({
           type: "done",
           reason: output.stopReason,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-provider-qoder-cn",
-  "version": "0.2.28",
+  "version": "0.2.30",
   "description": "Pi extension for Qoder CN — PAT/OAuth auth, /qoder-cn.usage quota command, and enterprise VPC endpoint support",
   "type": "module",
   "license": "MIT",