fix(cache): enable prompt caching for Qwen on OpenCode/OpenCode-Go/Alibaba (#13528)

Qwen models on OpenCode, OpenCode Go, and direct DashScope accept
Anthropic-style cache_control markers on OpenAI-wire chat completions,
but hermes only injected markers for Claude-named models. Result: zero
cache hits on every turn, full prompt re-billed — a community user
reported burning through their OpenCode Go subscription on Qwen3.6.

Extend _anthropic_prompt_cache_policy to return (True, False) — envelope
layout, not native — for the Alibaba provider family when the model name
contains 'qwen'. Envelope layout places markers on inner content blocks
(matching pi-mono's 'alibaba' cacheControlFormat) and correctly skips
top-level markers on tool-role messages (which OpenCode rejects).

Non-Qwen models on these providers (GLM, Kimi) keep their existing
behaviour — they have automatic server-side caching and don't need
client markers.

Upstream reference: pi-mono #3392 / #3393 documented this contract for
opencode-go Qwen models.

Adds 7 regression tests covering Qwen3.5/3.6/coder on each affected
provider plus negative cases for GLM/Kimi/OpenRouter-Qwen.

run_agent.py

@@ -2355,6 +2355,13 @@ class AIAgent:
         cost reduction as direct Anthropic callers, provided their
         gateway implements the Anthropic cache_control contract
         (MiniMax, Zhipu GLM, LiteLLM's Anthropic proxy mode all do).
+
+        Qwen / Alibaba-family models on OpenCode, OpenCode Go, and direct
+        Alibaba (DashScope) also honour Anthropic-style ``cache_control``
+        markers on OpenAI-wire chat completions. Upstream pi-mono #3392 /
+        pi #3393 documented this for opencode-go Qwen. Without markers
+        these providers serve zero cache hits, re-billing the full prompt
+        on every turn.
         """
         eff_provider = (provider if provider is not None else self.provider) or ""
         eff_base_url = base_url if base_url is not None else (self.base_url or "")
@@ -2362,7 +2369,9 @@ class AIAgent:
         eff_model = (model if model is not None else self.model) or ""
 
         base_lower = eff_base_url.lower()
-        is_claude = "claude" in eff_model.lower()
+        model_lower = eff_model.lower()
+        provider_lower = eff_provider.lower()
+        is_claude = "claude" in model_lower
         is_openrouter = base_url_host_matches(eff_base_url, "openrouter.ai")
         is_anthropic_wire = eff_api_mode == "anthropic_messages"
         is_native_anthropic = (
@@ -2377,6 +2386,22 @@ class AIAgent:
         if is_anthropic_wire and is_claude:
             # Third-party Anthropic-compatible gateway.
             return True, True
+
+        # Qwen/Alibaba on OpenCode (Zen/Go) and native DashScope: OpenAI-wire
+        # transport that accepts Anthropic-style cache_control markers and
+        # rewards them with real cache hits.  Without this branch
+        # qwen3.6-plus on opencode-go reports 0% cached tokens and burns
+        # through the subscription on every turn.
+        model_is_qwen = "qwen" in model_lower
+        provider_is_alibaba_family = provider_lower in {
+            "opencode", "opencode-zen", "opencode-go", "alibaba",
+        }
+        if provider_is_alibaba_family and model_is_qwen:
+            # Envelope layout (native_anthropic=False): markers on inner
+            # content parts, not top-level tool messages.  Matches
+            # pi-mono's "alibaba" cacheControlFormat.
+            return True, False
+
         return False, False
 
     @staticmethod

tests/run_agent/test_anthropic_prompt_cache_policy.py

@@ -118,6 +118,86 @@ class TestOpenAIWireFormatOnCustomProvider:
         assert agent._anthropic_prompt_cache_policy() == (False, False)
 
 
+class TestQwenAlibabaFamily:
+    """Qwen on OpenCode/OpenCode-Go/Alibaba — needs cache_control even on OpenAI-wire.
+
+    Upstream pi-mono #3392 / #3393 documented that these providers serve
+    zero cache hits without Anthropic-style markers. Regression reported
+    by community user (Qwen3.6 on opencode-go burning through
+    subscription with no cache). Envelope layout, not native, because the
+    wire format is OpenAI chat.completions.
+    """
+
+    def test_qwen_on_opencode_go_caches_with_envelope_layout(self):
+        agent = _make_agent(
+            provider="opencode-go",
+            base_url="https://opencode.ai/v1",
+            api_mode="chat_completions",
+            model="qwen3.6-plus",
+        )
+        should, native = agent._anthropic_prompt_cache_policy()
+        assert should is True, "Qwen on opencode-go must cache"
+        assert native is False, "opencode-go is OpenAI-wire; envelope layout"
+
+    def test_qwen35_plus_on_opencode_go(self):
+        agent = _make_agent(
+            provider="opencode-go",
+            base_url="https://opencode.ai/v1",
+            api_mode="chat_completions",
+            model="qwen3.5-plus",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (True, False)
+
+    def test_qwen_on_opencode_zen_caches(self):
+        agent = _make_agent(
+            provider="opencode",
+            base_url="https://opencode.ai/v1",
+            api_mode="chat_completions",
+            model="qwen3-coder-plus",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (True, False)
+
+    def test_qwen_on_direct_alibaba_caches(self):
+        agent = _make_agent(
+            provider="alibaba",
+            base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+            api_mode="chat_completions",
+            model="qwen3-coder",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (True, False)
+
+    def test_non_qwen_on_opencode_go_does_not_cache(self):
+        # GLM / Kimi on opencode-go don't need markers (they have automatic
+        # server-side caching or none at all).
+        agent = _make_agent(
+            provider="opencode-go",
+            base_url="https://opencode.ai/v1",
+            api_mode="chat_completions",
+            model="glm-5",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (False, False)
+
+    def test_kimi_on_opencode_go_does_not_cache(self):
+        agent = _make_agent(
+            provider="opencode-go",
+            base_url="https://opencode.ai/v1",
+            api_mode="chat_completions",
+            model="kimi-k2.5",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (False, False)
+
+    def test_qwen_on_openrouter_not_affected(self):
+        # Qwen via OpenRouter falls through — OpenRouter has its own
+        # upstream caching arrangement for Qwen (provider-dependent).
+        agent = _make_agent(
+            provider="openrouter",
+            base_url="https://openrouter.ai/api/v1",
+            api_mode="chat_completions",
+            model="qwen/qwen3-coder",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (False, False)
+
+
 class TestExplicitOverrides:
     """Policy accepts keyword overrides for switch_model / fallback activation."""