fix(opencode-go): send the session header Go now requires, add DeepSeek V4.1 Flash

docs/opencode-go.md

@@ -19,6 +19,8 @@ The provider uses one credential with two upstream wire protocols:
 
 The selective proxy diverts only explicit `clodex:opencode-go:...` model ids or saved aliases. Ordinary Claude model traffic remains on Claude Code's native Anthropic connection.
 
+Both endpoints require an `x-opencode-session` header and answer `MissingSessionID` without one. Clodex sends Claude Code's own session id on every Go request — proxy mode, the API server's `/anthropic/v1/messages`, and both branches of its `/openai/v1/chat/completions` — so Go can keep a conversation on one backend. A request with no client session id gets one stable per-process id instead. Note what that fallback means for a shared `clodex server`: every session-less client of one server process presents the same id to Go, which lets Go correlate those requests with each other. It discloses no content, and a header is required either way, so the alternative would be a fresh id per request and no prefix-cache locality at all.
+
 ## Supported transport scope
 
 The upstream OpenCode Go catalog also contains Responses-API models. Clodex intentionally excludes those entries from this provider (currently Grok and mainline GPT; other unmapped ids are reported by the updater until their transport is verified). The supported catalog contains only Anthropic Messages and Chat Completions models.

scripts/update-opencode-go-models.mjs

@@ -37,6 +37,8 @@ const ANTHROPIC_BASE_URL = 'https://opencode.ai/zen/go';
 // models (grok, mainline gpt) are deliberately absent.
 const TRANSPORTS = Object.assign(Object.create(null), {
   'deepseek-v4-flash': 'openai-completions',
+  // Measured 2026-09-11: V4.1 Flash answers on /v1/messages (thinking block + text).
+  'deepseek-v4.1-flash': 'anthropic-messages',
   'deepseek-v4-pro': 'openai-completions',
   'glm-5.1': 'openai-completions',
   'glm-5.2': 'openai-completions',

src/data/opencode-go-models.json

@@ -1,12 +1,12 @@
 [
   {
     "id": "deepseek-v4-flash",
-    "name": "DeepSeek V4 Flash (2x usage)",
+    "name": "DeepSeek V4 Flash",
     "contextWindow": 1000000,
     "cost": {
-      "input": 0.07,
-      "output": 0.14,
-      "cache_read": 0.0014
+      "input": 0.15,
+      "output": 0.6,
+      "cache_read": 0.003
     },
     "modelFormat": "openai",
     "npm": "@ai-sdk/openai-compatible",
@@ -34,12 +34,12 @@
   },
   {
     "id": "deepseek-v4-pro",
-    "name": "DeepSeek V4 Pro",
+    "name": "DeepSeek V4 Pro (New)",
     "contextWindow": 1000000,
     "cost": {
-      "input": 0.435,
-      "output": 0.87,
-      "cache_read": 0.003625
+      "input": 0.66,
+      "output": 1.98,
+      "cache_read": 0.022
     },
     "modelFormat": "openai",
     "npm": "@ai-sdk/openai-compatible",
@@ -65,6 +65,30 @@
     "upstreamModelId": "deepseek-v4-pro",
     "family": "deepseek"
   },
+  {
+    "id": "deepseek-v4.1-flash",
+    "name": "DeepSeek V4.1 Flash",
+    "contextWindow": 1000000,
+    "cost": {
+      "input": 0.15,
+      "output": 0.6,
+      "cache_read": 0.003
+    },
+    "modelFormat": "anthropic",
+    "npm": "@ai-sdk/anthropic",
+    "apiUrl": "https://opencode.ai/zen/go",
+    "reasoning": true,
+    "modalities": [
+      "text",
+      "image"
+    ],
+    "compatibility": {
+      "supportsCountTokens": false,
+      "supportsReasoningEffort": false
+    },
+    "upstreamModelId": "deepseek-v4.1-flash",
+    "family": "deepseek"
+  },
   {
     "id": "glm-5.1",
     "name": "GLM-5.1",
@@ -82,10 +106,10 @@
       "text"
     ],
     "compatibility": {
+      "supportsReasoningEffort": false,
       "supportsStore": false,
       "supportsDeveloperRole": false,
-      "maxTokensField": "max_tokens",
-      "supportsReasoningEffort": false
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "glm-5.1",
     "family": "glm"
@@ -125,13 +149,13 @@
   },
   {
     "id": "gpt-5.6-luna",
-    "name": "GPT-5.6 Luna (2x usage)",
+    "name": "GPT-5.6 Luna",
     "contextWindow": 1050000,
     "cost": {
-      "input": 0.1,
-      "output": 0.6,
-      "cache_read": 0.01,
-      "cache_write": 0.125
+      "input": 0.2,
+      "output": 1.2,
+      "cache_read": 0.02,
+      "cache_write": 0.25
     },
     "modelFormat": "openai",
     "npm": "@ai-sdk/openai-compatible",
@@ -141,8 +165,6 @@
       "text",
       "image"
     ],
-    "upstreamModelId": "gpt-5.6-luna",
-    "family": "gpt",
     "compatibility": {
       "reasoningEffortMap": {
         "off": "none",
@@ -156,7 +178,9 @@
       "supportsStore": false,
       "supportsDeveloperRole": false,
       "maxTokensField": "max_tokens"
-    }
+    },
+    "upstreamModelId": "gpt-5.6-luna",
+    "family": "gpt"
   },
   {
     "id": "hy3",
@@ -242,10 +266,10 @@
       "image"
     ],
     "compatibility": {
+      "supportsReasoningEffort": false,
       "supportsStore": false,
       "supportsDeveloperRole": false,
-      "maxTokensField": "max_tokens",
-      "supportsReasoningEffort": false
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "kimi-k2.7-code",
     "family": "kimi"
@@ -302,10 +326,10 @@
       "image"
     ],
     "compatibility": {
+      "supportsReasoningEffort": false,
       "supportsStore": false,
       "supportsDeveloperRole": false,
-      "maxTokensField": "max_tokens",
-      "supportsReasoningEffort": false
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "mimo-v2.5",
     "family": "mimo"
@@ -327,10 +351,10 @@
       "text"
     ],
     "compatibility": {
+      "supportsReasoningEffort": false,
       "supportsStore": false,
       "supportsDeveloperRole": false,
-      "maxTokensField": "max_tokens",
-      "supportsReasoningEffort": false
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "mimo-v2.5-pro",
     "family": "mimo"
@@ -352,10 +376,10 @@
       "text"
     ],
     "compatibility": {
+      "supportsReasoningEffort": false,
       "supportsStore": false,
       "supportsDeveloperRole": false,
-      "maxTokensField": "max_tokens",
-      "supportsReasoningEffort": false
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "minimax-m2.7",
     "family": "minimax"
@@ -377,12 +401,12 @@
       "text",
       "image"
     ],
-    "upstreamModelId": "minimax-m3",
-    "family": "minimax",
     "compatibility": {
       "supportsCountTokens": false,
       "supportsReasoningEffort": false
-    }
+    },
+    "upstreamModelId": "minimax-m3",
+    "family": "minimax"
   },
   {
     "id": "qwen3.6-plus",
@@ -403,10 +427,6 @@
       "image"
     ],
     "compatibility": {
-      "supportsStore": false,
-      "supportsDeveloperRole": false,
-      "thinkingFormat": "qwen",
-      "maxTokensField": "max_tokens",
       "reasoningEffortMap": {
         "low": "low",
         "medium": "medium",
@@ -415,7 +435,11 @@
         "max": "max",
         "minimal": null,
         "off": null
-      }
+      },
+      "supportsStore": false,
+      "supportsDeveloperRole": false,
+      "thinkingFormat": "qwen",
+      "maxTokensField": "max_tokens"
     },
     "upstreamModelId": "qwen3.6-plus",
     "family": "qwen3.6"
@@ -437,12 +461,12 @@
     "modalities": [
       "text"
     ],
-    "upstreamModelId": "qwen3.7-max",
-    "family": "qwen3.7",
     "compatibility": {
       "supportsCountTokens": false,
       "supportsReasoningEffort": false
-    }
+    },
+    "upstreamModelId": "qwen3.7-max",
+    "family": "qwen3.7"
   },
   {
     "id": "qwen3.7-plus",
@@ -462,12 +486,12 @@
       "text",
       "image"
     ],
-    "upstreamModelId": "qwen3.7-plus",
-    "family": "qwen3.7",
     "compatibility": {
       "supportsCountTokens": false,
       "supportsReasoningEffort": false
-    }
+    },
+    "upstreamModelId": "qwen3.7-plus",
+    "family": "qwen3.7"
   },
   {
     "id": "qwen3.8-max",
@@ -487,11 +511,11 @@
       "text",
       "image"
     ],
-    "upstreamModelId": "qwen3.8-max",
-    "family": "qwen3.8",
     "compatibility": {
       "supportsCountTokens": false,
       "supportsReasoningEffort": false
-    }
+    },
+    "upstreamModelId": "qwen3.8-max",
+    "family": "qwen3.8"
   }
 ]

src/data/opencode-go-models.ts

@@ -1,3 +1,4 @@
+import { randomUUID } from 'node:crypto';
 import models from './opencode-go-models.json';
 import type { CachedModel } from '../registry/types.js';
 
@@ -6,7 +7,7 @@ export const OPENCODE_GO_PROVIDER_NAME = 'OpenCode Go';
 export const OPENCODE_GO_COMPLETIONS_BASE_URL = 'https://opencode.ai/zen/go/v1';
 export const OPENCODE_GO_ANTHROPIC_BASE_URL = 'https://opencode.ai/zen/go';
 export const OPENCODE_GO_SOURCE = 'https://models.dev/api.json';
-export const OPENCODE_GO_SOURCE_FETCHED_AT = '2026-08-08T20:10:24.208Z';
+export const OPENCODE_GO_SOURCE_FETCHED_AT = '2026-09-11T08:20:44.219Z';
 
 type OpenCodeGoModel = Pick<CachedModel, 'id' | 'name'>
   & Partial<Omit<CachedModel, 'id' | 'name'>>;
@@ -25,3 +26,35 @@ type OpenCodeGoModel = Pick<CachedModel, 'id' | 'name'>
 export function buildOpenCodeGoModels(): OpenCodeGoModel[] {
   return structuredClone(models) as unknown as OpenCodeGoModel[];
 }
+
+/** Used when the client sent no session id; one per process, never per request. */
+const OPENCODE_GO_FALLBACK_SESSION_ID = randomUUID();
+
+/**
+ * OpenCode Go rejects any request that carries no session id — both
+ * `/v1/messages` and `/v1/chat/completions` answer
+ * `MissingSessionID: Request is missing x-opencode-session` (measured
+ * 2026-09-11 against the live endpoint). The header lets Go pin a conversation
+ * to one backend and keep its prefix cache warm, so forward Claude Code's own
+ * session id where one is known and fall back to a stable per-process id.
+ *
+ * A model counts as Go by provider id or by pointing at the Go base URL. The
+ * URL arm is what covers an imported or migrated provider whose id has drifted
+ * from the canonical `opencode-go` (a shape `registry/resolve-template.ts`
+ * supports): there the URL is the only remaining signal. It therefore has to
+ * read every field a Go entry can carry its URL in — `apiBaseUrl` is the one
+ * the runtime `ServerModelInfo` actually populates for openai-compatible
+ * providers (`server/models.ts`, built in `provider-catalog.ts`), and
+ * `baseUrl` is the anthropic-format sibling; `apiUrl` is the cached-registry
+ * spelling.
+ */
+export function openCodeGoSessionHeaders(
+  model: { providerId?: string; apiUrl?: string; baseUrl?: string; apiBaseUrl?: string },
+  claudeSessionId: string | undefined,
+): Record<string, string> | undefined {
+  const url = String(model.apiBaseUrl ?? model.baseUrl ?? model.apiUrl ?? '');
+  const isGo = model.providerId === OPENCODE_GO_PROVIDER_ID
+    || /^https:\/\/opencode\.ai\/zen\/go(\/|$)/.test(url);
+  if (!isGo) return undefined;
+  return { 'x-opencode-session': claudeSessionId ?? OPENCODE_GO_FALLBACK_SESSION_ID };
+}

src/provider-templates.ts

@@ -7,6 +7,7 @@ import {
   OPENCODE_GO_COMPLETIONS_BASE_URL,
   OPENCODE_GO_PROVIDER_ID,
   OPENCODE_GO_PROVIDER_NAME,
+  openCodeGoSessionHeaders,
 } from './data/opencode-go-models.js';
 
 export type ProviderAuthType = 'api' | 'oauth' | 'none';
@@ -81,6 +82,11 @@ export async function verifyOpenCodeGoCredential(apiKey: string): Promise<string
       headers: {
         'Content-Type': 'application/json',
         Authorization: `Bearer ${apiKey}`,
+        // Go answers `MissingSessionID` to any request without a session header,
+        // and whether it validates the key BEFORE that check is not something we
+        // have observed. Sending the header removes the question: the probe then
+        // cannot mistake a missing-session rejection for a verdict on the key.
+        ...openCodeGoSessionHeaders({ providerId: OPENCODE_GO_PROVIDER_ID }, undefined),
       },
       body: JSON.stringify({ model: model.upstreamModelId ?? model.id }),
       // This is the FIRST thing a user hits after pasting a key, and it runs

src/proxy.ts

@@ -59,6 +59,7 @@ import {
   estimateAnthropicInputTokens,
 } from './anthropic-endpoints.js';
 import { withResponsesWebSocketDiagnosticContext } from './oauth/responses-websocket.js';
+import { openCodeGoSessionHeaders } from './data/opencode-go-models.js';
 import { resolveContextWindow } from './context-window.js';
 import { listenTcpServer } from './listener-ready.js';
 import type { ModelRuntimeCompatibility } from './model-runtime-compatibility.js';
@@ -537,6 +538,14 @@ export async function startProxyCatalog(
 
         let effectiveBeta = inboundBeta;
         let claudeCodeSessionId: string | undefined;
+        const inboundSessionRaw = req.headers['x-claude-code-session-id'];
+        const goSessionHeaders = openCodeGoSessionHeaders(
+          { providerId: route.providerId, baseUrl: upstreamUrl },
+          extractClaudeSessionId(
+            anthropicBody,
+            Array.isArray(inboundSessionRaw) ? inboundSessionRaw[0] : inboundSessionRaw,
+          ),
+        );
         if (isOAuth) {
           // Identity injection and beta selection for Claude Code OAuth.
           const seed = route.providerId ?? route.realModelId;
@@ -556,7 +565,7 @@ export async function startProxyCatalog(
             authType: routeAuthType,
             log: message => plog(message),
             claudeCodeSessionId,
-            extraHeaders: route.headers,
+            extraHeaders: { ...route.headers, ...goSessionHeaders },
             refreshToken: route.refreshToken,
             onTokenRefreshed: refreshed => { route.apiKey = refreshed; },
             signal: clientAbort.signal,
@@ -627,6 +636,11 @@ export async function startProxyCatalog(
               upstreamModelId: route.realModelId,
             },
           });
+          const goSdkSessionHeaders = openCodeGoSessionHeaders(
+            { providerId: route.providerId, baseUrl: route.baseURL },
+            claudeSessionId,
+          );
+          if (goSdkSessionHeaders) params.headers = { ...params.headers, ...goSdkSessionHeaders };
           plog(() =>
             `sdk: npm=${route.npm} model=${route.realModelId}, stream=${clientWantsStream}, ` +
             `tools=${anthropicBody.tools?.length ?? 0}, msgs=${params.messages.length}`,

src/sdk-adapter.ts

@@ -202,6 +202,8 @@ export interface SdkCallParams {
   maxOutputTokens?: number;
   temperature?: number;
   providerOptions?: Record<string, Record<string, unknown>>;
+  /** Per-request upstream headers; `streamText`/`generateText` take them as-is. */
+  headers?: Record<string, string>;
 }
 
 // ── system ───────────────────────────────────────────────────────────────────