[Feat] Add Qwen3.8 Max to OpenCode Go

packages/types/src/__tests__/lite-llm.test.ts

@@ -21,6 +21,7 @@ describe("LiteLLM preserveReasoning model detection", () => {
 	it("matches case-insensitively", () => {
 		expect(isLiteLLMPreserveReasoningModel("MiniMax-M2.7-Highspeed")).toBe(true)
 		expect(isLiteLLMPreserveReasoningModel("GLM-5.3")).toBe(true)
+		expect(isLiteLLMPreserveReasoningModel("QWEN3.8-MAX")).toBe(true)
 	})
 
 	it("does not match model ids that merely contain a known family as a substring", () => {

packages/types/src/__tests__/opencode-go.test.ts

@@ -10,6 +10,7 @@ import {
 
 describe("opencode-go registry", () => {
 	const anthropicFormatModels = [
+		"qwen3.8-max",
 		"qwen3.7-max",
 		"qwen3.7-plus",
 		"qwen3.6-plus",
@@ -80,6 +81,22 @@ describe("opencode-go registry", () => {
 			expect(info?.cacheReadsPrice).toBe(0.3)
 		})
 
+		it("exposes current Qwen3.8 Max capabilities and Go pricing", () => {
+			const info = getOpencodeGoModelInfo("qwen3.8-max")
+			expect(info).toMatchObject({
+				maxTokens: 131_072,
+				contextWindow: 1_000_000,
+				supportsImages: true,
+				supportsPromptCache: true,
+				supportsMaxTokens: true,
+				inputPrice: 2.0,
+				outputPrice: 6.0,
+				cacheReadsPrice: 0.25,
+				cacheWritesPrice: 2.5,
+			})
+			expect(info?.preserveReasoning).toBeUndefined()
+		})
+
 		it("glm-5.3 exposes its native context, pricing, and always-on reasoning levels", () => {
 			const info = getOpencodeGoModelInfo("glm-5.3")
 			expect(info).toBeDefined()

packages/types/src/__tests__/provider-settings.test.ts

@@ -130,6 +130,7 @@ describe("getApiProtocol", () => {
 
 	describe("Opencode Go provider", () => {
 		it("should return 'anthropic' for opencode-go Anthropic-format models (Qwen/MiniMax)", () => {
+			expect(getApiProtocol(providerIdentifiers.opencodeGo, "qwen3.8-max")).toBe(ANTHROPIC_API_PROTOCOL)
 			expect(getApiProtocol(providerIdentifiers.opencodeGo, "qwen3.7-max")).toBe(ANTHROPIC_API_PROTOCOL)
 			expect(getApiProtocol(providerIdentifiers.opencodeGo, "qwen3.7-plus")).toBe(ANTHROPIC_API_PROTOCOL)
 			expect(getApiProtocol(providerIdentifiers.opencodeGo, "qwen3.6-plus")).toBe(ANTHROPIC_API_PROTOCOL)

packages/types/src/providers/lite-llm.ts

@@ -24,14 +24,12 @@ export const litellmDefaultModelInfo: ModelInfo = {
  *
  * Rather than matching model-family substrings with a regex (which can
  * over-match unrelated aliases, e.g. a family fragment appearing inside a
- * longer unrelated model id), this is an explicit list of the exact model
- * ids that set `preserveReasoning: true` in their native provider config
- * (see deepseek.ts, mimo.ts, moonshot.ts, bedrock.ts, fireworks.ts, zai.ts,
- * minimax.ts, opencode-go.ts). The same behavior is inferred for a
- * LiteLLM-routed alias of the same underlying model. Keep this list in sync
- * with those registries. This is still best-effort: unrecognized aliases or
- * renamed deployments will not match, and callers should treat it as a
- * heuristic, not a source of truth.
+ * longer unrelated model id), this is an explicit list of models whose
+ * OpenAI-compatible routes use interleaved `reasoning_content`. Native
+ * provider metadata informs this list where applicable, but gateway routes
+ * can have different preservation semantics. This is still best-effort:
+ * unrecognized aliases or renamed deployments will not match, and callers
+ * should treat it as a heuristic, not a source of truth.
  */
 export const LITELLM_PRESERVE_REASONING_MODEL_IDS = [
 	// deepseek.ts
@@ -75,6 +73,7 @@ export const LITELLM_PRESERVE_REASONING_MODEL_IDS = [
 	"qwen3.6-plus",
 	"qwen3.7-plus",
 	"qwen3.7-max",
+	"qwen3.8-max",
 ] as const
 
 const LITELLM_PRESERVE_REASONING_MODEL_ID_SET = new Set<string>(LITELLM_PRESERVE_REASONING_MODEL_IDS)

packages/types/src/providers/opencode-go.ts

@@ -307,6 +307,19 @@ export const opencodeGoModels: Record<string, ModelInfo> = {
 		description:
 			"Qwen3.7 Max - Alibaba's flagship text-only reasoning agent model with a 1M context window, designed for long-horizon agent workflows. Available via the Opencode Go plan.",
 	},
+	"qwen3.8-max": {
+		maxTokens: 131_072,
+		contextWindow: 1_000_000,
+		supportsImages: true,
+		supportsPromptCache: true,
+		supportsMaxTokens: true,
+		inputPrice: 2.0,
+		outputPrice: 6.0,
+		cacheReadsPrice: 0.25,
+		cacheWritesPrice: 2.5,
+		description:
+			"Qwen3.8 Max - Alibaba's flagship multimodal reasoning model with a 1M context window, 128k max output, and long-horizon coding and agentic capabilities. Available via the Opencode Go plan.",
+	},
 
 	// --- DeepSeek ---
 	"deepseek-v4-pro": {
@@ -364,6 +377,7 @@ export const opencodeGoModels: Record<string, ModelInfo> = {
  */
 export const OPENCODE_GO_ANTHROPIC_FORMAT_MODELS = new Set<string>([
 	// --- Alibaba Qwen ---
+	"qwen3.8-max",
 	"qwen3.7-max",
 	"qwen3.7-plus",
 	"qwen3.6-plus",

src/api/providers/opencode-go.ts

@@ -54,9 +54,9 @@ import {
  *
  *   - OpenAI-compatible chat completions (`/v1/chat/completions`, "oa-compat")
  *     — used by GLM, Kimi, DeepSeek, and MiMo models.
- *   - Anthropic Messages (`/v1/messages`) — used by Qwen (qwen3.7-max,
- *     qwen3.7-plus, qwen3.6-plus) and MiniMax (minimax-m3, minimax-m2.7,
- *     minimax-m2.5) models.
+ *   - Anthropic Messages (`/v1/messages`) — used by Qwen (qwen3.8-max,
+ *     qwen3.7-max, qwen3.7-plus, qwen3.6-plus) and MiniMax (minimax-m3,
+ *     minimax-m2.7, minimax-m2.5) models.
  *
  * Sending an Anthropic-format model to the chat completions endpoint is
  * rejected with `401 Model <id> is not supported for format oa-compat`, so this