ltmoerdani · Oct 20, 2026 · Oct 15, 2026
diff --git a/ARCHITECTURE-MAP.md b/ARCHITECTURE-MAP.md
@@ -234,7 +234,7 @@ flowchart TD
 | Claude        | `/^claude-/i`                                             | `messages`         | `null`         | any                                        |
 | MiniMax m2.x  | `/^minimax-m2\./i`                                        | `messages`         | `minimax`      | `opencodego` only (Zen → chat-completions) |
 | Qwen Messages | `/^qwen3\.(?:5\|6)-plus(?:-free)?$/` · `/^qwen3\.7-max$/` | `messages`         | `qwen`         | any                                        |
-| Gemini        | `/^gemini-/i`                                             | `google`           | `null`         | `opencodezen` only                         |
+| Gemini        | `/^gemini-/i`                                             | `google`           | `gemini`       | any                                        |
 | MiniMax       | `/^minimax-/i`                                            | `chat-completions` | `minimax`      | any                                        |
 | DeepSeek      | `/^deepseek-/i`                                           | `chat-completions` | `deepseek`     | any                                        |
 | GLM           | `/^glm-/i`                                                | `chat-completions` | `glm`          | any                                        |
@@ -245,7 +245,7 @@ flowchart TD
 
 > **Scope note:** context limits / capabilities / cost are **NOT** in this table — they stay metadata-driven (live models.dev via `src/models/metadata.ts`). Do not duplicate them into a static table.
 
-**Endpoint URL resolution** (`resolveModelRouting`): agent-host variants resolve to their base vendor first; `responses` → `provider.responsesUrl`, `messages` → `provider.messagesUrl`, `google` → `${provider.modelsUrl}/${modelId}`, default → `provider.chatCompletionsUrl`.
+**Endpoint URL resolution** (`resolveModelRouting`): agent-host variants resolve to their base vendor first; `responses` → `provider.responsesUrl`, `messages` → `provider.messagesUrl`, `google` → `${provider.googleModelsUrl ?? provider.modelsUrl}/${modelId}`, default → `provider.chatCompletionsUrl`.
 
 ---
 
@@ -476,7 +476,7 @@ On push/PR to `main`/`develop`, Node 20: `npm ci` → `compile` → `lint` (with
 - **Activation:** `onStartupFinished`, `onLanguageModelChatProvider:opencodego`, `onLanguageModelChatProvider:opencodezen`
 - **Contribution points:** `commands` (16), `configuration` (35+ keys), `languageModelChatProviders` (4 vendors: `opencodego`, `opencodezen`, `opencodego-agent`, `opencodezen-agent`)
 - **4 providers:** Go (paid) + Zen (free default) + agent-host variants mirroring each base vendor
-- **Endpoints:** `https://opencode.ai/zen/{go/,}v1/{models,chat/completions,messages,responses}` + `https://opencode.ai/zen/go/v1/usage` (server meters) + `https://models.dev/api.json` (metadata)
+- **Endpoints:** `https://opencode.ai/zen/{go/,}v1/{models,chat/completions,messages,responses}` + `https://opencode.ai/inference/go/v1/models` (Go Gemini route, see `googleModelsUrl`) + `https://opencode.ai/zen/go/v1/usage` (server meters) + `https://models.dev/api.json` (metadata)
 
 ---
 

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,20 @@
 
 All notable changes to the **OpenCode Go BYOK Provider** extension are documented here.
 
+## [0.7.10] — unreleased
+
+### Added
+
+- **`[Thinking]` Gemini models now get a Thinking Effort picker.** Gemini had no thinking control at all: its registry row declared no thinking family, and the generic `reasoning_effort` fallback is deliberately gated to the chat-completions endpoint (that field is meaningless on the Google wire format), so the picker never rendered and the request went out with `thinkingPayload={}` — the model always answered at its default effort. A new `GeminiThinking` strategy maps the pick to Google's native `generationConfig.thinkingConfig` (`thinkingLevel` + `includeThoughts: true`); verified live on `gemini-4-argon-contributor` that `thinkingLevel: "medium"/"high"` is accepted (HTTP 200) and thinking is actually performed (`usageMetadata.thoughtsTokenCount` billed). Caveat documented in `thinking/gemini.ts`: the gateway relays only an opaque `thoughtSignature`, not thought summaries, so the thinking panel stays empty for Gemini (same class as Muse Spark's `encrypted_content`) — if upstream starts relaying `thought: true` parts the existing Google normalizer maps them with no further client work. Levels come from the model's own models.dev `reasoning_options` (`off`/`on`/`low`/`medium`/`high`), with `opencodego.thinking.gemini` as the workspace default. `off` sends no thinking config at all (model default, unchanged from before).
+
+### Fixed
+
+- **`[Models]` Gemini models served by OpenCode Go now route to the Google API instead of chat-completions.** Every request to `gemini-4-argon-contributor` failed with `400 Model does not support this protocol.` because the `gemini` row in `core/registry.ts` was restricted to `opencodezen`, so on Go the lookup fell through to the catch-all row and the model was sent to `chat/completions` — while the gateway serves it Google-native (models.dev marks it `@ai-sdk/google`, and OpenCode's own Go endpoint table lists it under `@ai-sdk/google` too). The row now applies to every vendor. Fixing that exposed the second half of the bug: the Google route is built as `{base}/{modelId}:streamGenerateContent?alt=sse`, and Go's built-in `/zen/go/v1` base does not serve that path at all (it answers with the opencode.ai website 404 page), so provider routing gained an optional `googleModelsUrl` — Go's built-in default points at `https://opencode.ai/inference/go/v1/models` (the base OpenCode's own endpoint table documents for the model), while Zen and any configured `apiBaseUrl` gateway keep deriving the route from `modelsUrl` exactly as before (issue #247 contract intact). Every other route (chat/completions, messages, responses, models, usage) is untouched. Routing tests now cover Go, Zen, and both agent-host variants. Registry drift found while investigating (Go/Zen `minimax-m3`, `qwen3.8-max`, `qwen3.8-flash`, `qwen3.7-plus` documented as Anthropic `messages`) is deliberately not changed here — each needs live verification first. Documented in `docs/features/17-20260814-data-driven-model-registry.md`.
+
+- **`[Usage]` Google-route responses now report usage and finish reasons.** `updateRequestUsageSummary` understood OpenAI, Responses, and Anthropic payload shapes but not Google's `usageMetadata` / `candidates[].finishReason`, so every Gemini request logged `promptTokens=n/a … finishReason=<unknown>` and the Go usage meter recorded `promptTokens=0` — silently losing cost tracking for the whole family. The Google normalizer's own usage mapping is now reused for the summary. With usage visible again, the empty-response diagnostic (`[diag-empty-response]`, the raw-SSE dump built for exactly the "gateway sent bytes but nothing parsed" case) no longer requires completion tokens to fire — that precondition made it unreachable on Google responses and left an empty `200 OK` stream of 5 unparsed events undiagnosable.
+
+- **`[Models]` Gemini requests no longer send `temperature` to models that declare it unsupported.** The Google body builder sent `temperature` unconditionally even when models.dev marks the model `temperature: false` (`gemini-4-argon-contributor`) — the OpenAI and Anthropic builders already guarded this. Guarded the same way, and the builder now receives the model metadata it needs for both this and the thinking payload.
+
 ## [0.7.9] — 2026-10-13
 
 ### Fixed

diff --git a/README.md b/README.md
@@ -198,7 +198,7 @@ Deprecated/unavailable models are filtered before registration. Per-provider lim
 | Family                                                | Endpoint                         | Why                  |
 | ----------------------------------------------------- | -------------------------------- | -------------------- |
 | Zen GPT (`gpt-*`)                                     | `/responses`                     | OpenAI native        |
-| Zen Gemini (`gemini-*`)                               | `:streamGenerateContent?alt=sse` | Google native        |
+| Gemini (`gemini-*`, Zen + Go)                         | `:streamGenerateContent?alt=sse` | Google native        |
 | Zen Claude (`claude-*`) + Go MiniMax (`minimax-m2.*`) | `/messages`                      | Anthropic-compatible |
 | Everything else (Qwen, DeepSeek, GLM, Kimi, MiMo…)    | `/chat/completions`              | OpenAI-compatible    |
 
@@ -248,6 +248,7 @@ Per-model reasoning configuration, dynamically enhanced with `reasoning_options`
 | **MiniMax**                    | `off` / `on`                                                                          | `opencodego.thinking.minimax`              |
 | **Mimo (Xiaomi)**              | `off` / `low` / `medium` / `high`                                                     | `opencodego.thinking.mimo`                 |
 | **Qwen**                       | `auto` / `on` / `off` + `thinking_budget` (4096–81920)                                | `opencodego.thinking.qwen` + `.qwenBudget` |
+| **Gemini (Google)**            | `off` / `on` / `low` / `medium` / `high` (`thinkingLevel`, thoughts streamed)         | `opencodego.thinking.gemini`               |
 | **Any future reasoning model** | `off` / `low` / `medium` / `high` / `xhigh` / `max` (auto-detected from `models.dev`) | `opencodego.thinking.generic`              |
 
 > **`opencodego.debugReasoning`** writes provider `reasoning_content` to **Output → OpenCode** for debugging.
@@ -414,6 +415,7 @@ Provider diagnostics also include the VS Code/extension versions, extension host
 | `opencodego.thinking.mimo`                | `"off"`                         | `off`/`low`/`medium`/`high`                                                                                            |
 | `opencodego.thinking.qwen`                | `"off"`                         | `auto`/`on`/`off`                                                                                                      |
 | `opencodego.thinking.qwenBudget`          | `"auto"`                        | `auto`/`4096`/`16384`/`32768`/`81920`                                                                                  |
+| `opencodego.thinking.gemini`              | `"off"`                         | `off`/`on`/`low`/`medium`/`high`                                                                                       |
 
 <details>
 <summary><b>📜 Full settings reference with descriptions</b></summary>

diff --git a/docs/architecture/01-20260514-open-code-provider-architecture.md b/docs/architecture/01-20260514-open-code-provider-architecture.md
@@ -202,7 +202,7 @@ Routing is centralized in `src/routing.ts`.
 | Claude family                      | `messages`         | `/zen/v1/messages`                          |
 | Go MiniMax M2 family               | `messages`         | `/zen/go/v1/messages`                       |
 | Qwen 3.5/3.6 Plus and Qwen 3.7 Max | `messages`         | provider messages endpoint                  |
-| Zen Gemini family                  | `google`           | `streamGenerateContent?alt=sse` style route |
+| Gemini family (Zen + Go)           | `google`           | `streamGenerateContent?alt=sse` style route |
 | All other models                   | `chat-completions` | provider chat-completions endpoint          |
 
 The request layer maps VS Code chat parts and tools into the correct request body for the selected endpoint.

diff --git a/docs/features/17-20260814-data-driven-model-registry.md b/docs/features/17-20260814-data-driven-model-registry.md
@@ -116,22 +116,22 @@ interface ModelRegistryEntry {
 
 ### Current rows (11 families + catch-all)
 
-| Order | Family          | Patterns                                                   | endpointKind     | thinkingFamily | vendors     |
-| ----- | --------------- | ---------------------------------------------------------- | ---------------- | -------------- | ----------- |
-| 1     | `gpt`           | `/^gpt-/i`                                                 | responses        | openai         | —           |
-| 2     | `claude`        | `/^claude-/i`                                              | messages         | null           | —           |
-| 3     | `minimax-m2`    | `/^minimax-m2\./i`                                         | messages         | minimax        | opencodego  |
-| 4     | `qwen-messages` | `/^qwen3\.(?:5\|6)-plus(?:-free)?$/i`, `/^qwen3\.7-max$/i` | messages         | qwen           | —           |
-| 5     | `gemini`        | `/^gemini-/i`                                              | google           | null           | opencodezen |
-| 6     | `minimax`       | `/^minimax-/i`                                             | chat-completions | minimax        | —           |
-| 7     | `deepseek`      | `/^deepseek-/i`                                            | chat-completions | deepseek       | —           |
-| 8     | `glm`           | `/^glm-/i`                                                 | chat-completions | glm            | —           |
-| 9     | `kimi`          | `/^kimi-/i`                                                | chat-completions | kimi           | —           |
-| 10    | `mimo`          | `/^mimo-/i`                                                | chat-completions | mimo           | —           |
-| 11    | `qwen`          | `/^qwen3(?:\.\|-)/i`                                       | chat-completions | qwen           | —           |
-| 12    | `default`       | `/.*/`                                                     | chat-completions | null           | —           |
-
-**Vendor-restriction semantics:** `minimax-m2.*` served by **Go** → Messages API (Anthropic); on **Zen** it falls through to the generic `minimax` row → chat-completions. `gemini-*` served by **Zen** → Google API; on Go → chat-completions. Agent-host variants are resolved to their base vendor via `resolveBaseVendor()` before lookup.
+| Order | Family          | Patterns                                                   | endpointKind     | thinkingFamily | vendors    |
+| ----- | --------------- | ---------------------------------------------------------- | ---------------- | -------------- | ---------- |
+| 1     | `gpt`           | `/^gpt-/i`                                                 | responses        | openai         | —          |
+| 2     | `claude`        | `/^claude-/i`                                              | messages         | null           | —          |
+| 3     | `minimax-m2`    | `/^minimax-m2\./i`                                         | messages         | minimax        | opencodego |
+| 4     | `qwen-messages` | `/^qwen3\.(?:5\|6)-plus(?:-free)?$/i`, `/^qwen3\.7-max$/i` | messages         | qwen           | —          |
+| 5     | `gemini`        | `/^gemini-/i`                                              | google           | gemini         | —          |
+| 6     | `minimax`       | `/^minimax-/i`                                             | chat-completions | minimax        | —          |
+| 7     | `deepseek`      | `/^deepseek-/i`                                            | chat-completions | deepseek       | —          |
+| 8     | `glm`           | `/^glm-/i`                                                 | chat-completions | glm            | —          |
+| 9     | `kimi`          | `/^kimi-/i`                                                | chat-completions | kimi           | —          |
+| 10    | `mimo`          | `/^mimo-/i`                                                | chat-completions | mimo           | —          |
+| 11    | `qwen`          | `/^qwen3(?:\.\|-)/i`                                       | chat-completions | qwen           | —          |
+| 12    | `default`       | `/.*/`                                                     | chat-completions | null           | —          |
+
+**Vendor-restriction semantics:** `minimax-m2.*` served by **Go** → Messages API (Anthropic); on **Zen** it falls through to the generic `minimax` row → chat-completions. `gemini-*` → Google API on **both** vendors; the route base is `googleModelsUrl ?? modelsUrl`, so Go's built-in gateway uses its own Google base while Zen (and any custom `apiBaseUrl` gateway) keeps building from `modelsUrl`. Agent-host variants are resolved to their base vendor via `resolveBaseVendor()` before lookup.
 
 ### How to add a model family
 
@@ -149,7 +149,7 @@ interface ModelRegistryEntry {
 | `claude-`                                                   | **messages** (Anthropic)             | any vendor                                                                             |
 | `minimax-m2.*` (Go)                                         | **messages**                         | on Zen → chat-completions                                                              |
 | `qwen3.5-plus(-free)`, `qwen3.6-plus(-free)`, `qwen3.7-max` | **messages**                         | others → chat-completions                                                              |
-| `gemini-*` (Zen)                                            | **google**                           | on Go → chat-completions                                                               |
+| `gemini-*` (Zen + Go)                                       | **google**                           | Go uses `googleModelsUrl` base, Zen uses `modelsUrl`                                   |
 | everything else                                             | **chat-completions**                 | default                                                                                |
 
 ---

diff --git a/package.json b/package.json
@@ -457,6 +457,18 @@
           "default": "off",
           "markdownDescription": "Thinking mode for Muse Spark models (`muse-spark-1.2-contributor`, `muse-spark-1.2-contributor-free`). Maps to the Responses API `reasoning.effort` field. Higher effort increases reasoning depth and may increase latency and usage."
         },
+        "opencodego.thinking.gemini": {
+          "type": "string",
+          "enum": [
+            "off",
+            "on",
+            "low",
+            "medium",
+            "high"
+          ],
+          "default": "off",
+          "markdownDescription": "Thinking level for Gemini models (Google API wire format). Maps to `generationConfig.thinkingConfig.thinkingLevel` and sets `includeThoughts: true` so reasoning streams into the thinking panel. `off` sends no thinking config (model default); `on` surfaces thoughts without pinning a level. Higher levels increase reasoning depth and latency."
+        },
         "opencodego.thinking.generic": {
           "type": "string",
           "enum": [

diff --git a/src/config.ts b/src/config.ts
@@ -61,6 +61,7 @@ export const SETTING_THINKING_QWEN = "thinking.qwen";
 export const SETTING_THINKING_QWEN_BUDGET = "thinking.qwenBudget";
 export const SETTING_THINKING_MIMO = "thinking.mimo";
 export const SETTING_THINKING_MUSE = "thinking.muse";
+export const SETTING_THINKING_GEMINI = "thinking.gemini";
 export const SETTING_THINKING_GENERIC = "thinking.generic";
 
 // ─── Inline completions (issue #49) ─────────────────────────────────────────
@@ -128,6 +129,15 @@ export const DEFAULT_MODEL_MAX_OUTPUT_TOKENS = 65536;
 
 /** Default OpenCode Go API base URL; can be overridden in VS Code settings. */
 export const DEFAULT_GO_API_BASE_URL = "https://opencode.ai/zen/go/v1";
+/**
+ * Base of the Google-native (Gemini) generate route on the built-in Go gateway.
+ * Go serves that route on a different base than the rest of its routes: the
+ * `/zen/go/v1/models/{id}:streamGenerateContent` path is not routed at all
+ * (404), while `/inference/go/v1/models/{id}` is (verified 2026-10-14). A
+ * configured `apiBaseUrl` drives every route including this one (issue #247),
+ * so only the built-in default carries the split (see `provider/definitions.ts`).
+ */
+export const DEFAULT_GO_GOOGLE_MODELS_URL = "https://opencode.ai/inference/go/v1/models";
 /** Default OpenCode Zen API base URL; can be overridden in VS Code settings. */
 export const DEFAULT_ZEN_API_BASE_URL = "https://opencode.ai/zen/v1";
 
@@ -426,5 +436,6 @@ export const THINKING_DEFAULTS = {
   qwenBudget: "auto",
   mimo: "off",
   muse: "off",
+  gemini: "off",
   generic: "off",
 } as const;