opencode: Support interleaved_reasoning and fix DeepSeek (#55574)

OpenCode API endpoints for DeepSeek were [moved from
Anthropic-compatible to
OpenAI-compatible](https://github.com/anomalyco/opencode/pull/24500) and
DeepSeek requires interleaved reasoning enabled to work. I ran a
_"rename this variable to potato"_ test and I can confirm DeepSeek V4
Flash and Pro both work now 🎉

Some other OpenCode Go models were marked [on
models.dev](https://github.com/anomalyco/models.dev/tree/dev/providers/opencode-go/models)
as supporting `interleaved_reasoning` so they too got that enabled. Kimi
K2.5 and Kimi K2.6 continue to fail with
https://github.com/zed-industries/zed/issues/51743
(https://github.com/zed-industries/zed/pull/55085 seems to hint at this
being [an OpenCode
issue](https://github.com/zed-industries/zed/issues/51743#issuecomment-4336785765)?),
but all other models seem to work fine both with `interleaved_reasoning`
and without it 🤷 I assume it's better to have that turned on? Again, the
intersection of OpenAI Chat Completions API, different models, different
inference providers, how they all work together is something I know
nothing about!

Self-Review Checklist:

- [X] I've reviewed my own diff for quality, security, and reliability
- [X] Unsafe blocks (if any) have justifying comments
- [X] The content is consistent with the [UI/UX
checklist](https://github.com/zed-industries/zed/blob/main/CONTRIBUTING.md#uiux-checklist)
- [ ] Tests cover the new/changed behavior
- [X] Performance impact has been considered and is acceptable

Release Notes:
- OpenCode Go: use correct DeepSeek endpoints
- OpenCode: add support for interleaved_reasoning

crates/language_models/src/provider/opencode.rs

@@ -278,6 +278,7 @@ impl LanguageModelProvider for OpenCodeLanguageModelProvider {
                 protocol,
                 reasoning_effort_levels: model.reasoning_effort_levels.clone(),
                 custom_model_api_url: model.custom_model_api_url.clone(),
+                interleaved_reasoning: model.interleaved_reasoning,
             };
             let key = format!("{}/{}", subscription.id_prefix(), model.name);
             models.insert(key, (custom_model, subscription));
@@ -664,7 +665,7 @@ impl LanguageModel for OpenCodeLanguageModel {
                     false,
                     self.model.max_output_tokens(),
                     reasoning_effort,
-                    false,
+                    self.model.interleaved_reasoning(),
                 );
                 let stream = self.stream_openai_chat(openai_request, http_client, cx);
                 async move {

crates/opencode/src/opencode.rs

@@ -168,6 +168,7 @@ pub enum Model {
         protocol: ApiProtocol,
         reasoning_effort_levels: Option<Vec<ReasoningEffort>>,
         custom_model_api_url: Option<String>,
+        interleaved_reasoning: bool,
     },
 }
 
@@ -385,8 +386,6 @@ impl Model {
 
             Self::Gemini3_1Pro | Self::Gemini3Flash => ApiProtocol::Google,
 
-            Self::DeepSeekV4Pro | Self::DeepSeekV4Flash => ApiProtocol::Anthropic,
-
             Self::MiniMaxM2_5Free
             | Self::Glm5
             | Self::Glm5_1
@@ -398,6 +397,8 @@ impl Model {
             | Self::MimoV2_5
             | Self::Qwen3_5Plus
             | Self::Qwen3_6Plus
+            | Self::DeepSeekV4Pro
+            | Self::DeepSeekV4Flash
             | Self::BigPickle
             | Self::Nemotron3SuperFree
             | Self::Ling2_6FlashFree
@@ -407,6 +408,27 @@ impl Model {
         }
     }
 
+    pub fn interleaved_reasoning(&self) -> bool {
+        match self {
+            Self::DeepSeekV4Pro
+            | Self::DeepSeekV4Flash
+            | Self::KimiK2_5
+            | Self::KimiK2_6
+            | Self::MimoV2Omni
+            | Self::MimoV2_5
+            | Self::MimoV2_5Pro
+            | Self::Glm5
+            | Self::Glm5_1 => true,
+
+            Self::Custom {
+                interleaved_reasoning,
+                ..
+            } => *interleaved_reasoning,
+
+            _ => false,
+        }
+    }
+
     pub fn max_token_count(&self) -> u64 {
         match self {
             // Anthropic models
@@ -487,9 +509,6 @@ impl Model {
             // Google models
             Self::Gemini3_1Pro | Self::Gemini3Flash => Some(65_536),
 
-            // Anthropic-compatible models
-            Self::DeepSeekV4Pro | Self::DeepSeekV4Flash => Some(384_000),
-
             // OpenAI-compatible models
             Self::MiniMaxM2_7 => Some(131_072),
             Self::MiniMaxM2_5 | Self::MiniMaxM2_5Free => Some(131_072),
@@ -497,6 +516,7 @@ impl Model {
             Self::BigPickle => Some(128_000),
             Self::KimiK2_6 | Self::KimiK2_5 => Some(65_536),
             Self::Qwen3_5Plus | Self::Qwen3_6Plus => Some(65_536),
+            Self::DeepSeekV4Pro | Self::DeepSeekV4Flash => Some(384_000),
             Self::Nemotron3SuperFree => Some(128_000),
             Self::MimoV2_5Pro | Self::MimoV2_5 | Self::MimoV2Pro | Self::MimoV2Omni => {
                 Some(128_000)
@@ -565,14 +585,13 @@ impl Model {
             | Self::MiniMaxM2_7
             | Self::MimoV2Pro
             | Self::MimoV2_5Pro
+            | Self::DeepSeekV4Pro
+            | Self::DeepSeekV4Flash
             | Self::BigPickle
             | Self::Nemotron3SuperFree
             | Self::Ling2_6FlashFree
             | Self::Hy3PreviewFree => false,
 
-            // DeepSeek models (Anthropic protocol) don't support images
-            Self::DeepSeekV4Pro | Self::DeepSeekV4Flash => false,
-
             Self::Custom { protocol, .. } => matches!(
                 protocol,
                 ApiProtocol::Anthropic

crates/settings_content/src/language_model.rs

@@ -181,6 +181,9 @@ pub struct OpenCodeAvailableModel {
     pub custom_model_api_url: Option<String>,
     /// Supported reasoning effort levels, for example `["low", "medium", "high"].
     pub reasoning_effort_levels: Option<Vec<ReasoningEffort>>,
+    /// When using OpenAiChat protocol, whether thinking tokens are sent as a dedicated `reasoning_content` field or inline in message text.
+    #[serde(default)]
+    pub interleaved_reasoning: bool,
 }
 
 #[with_fallible_options]

docs/src/ai/llm-providers.md

@@ -663,6 +663,7 @@ The Zed agent comes pre-configured with OpenCode models. If you wish to use newe
           "max_output_tokens": 98765,
           "protocol": "openai_chat",
           "reasoning_effort_levels": ["low", "medium", "high"],
+          "interleaved_reasoning": false,
           "subscription": "go",
           "custom_model_api_url": "https://example.com/zen"
         }
@@ -680,6 +681,7 @@ The available configuration options for custom models are:
 - `max_output_tokens` (optional): maximum tokens the model can generate, for example `64000`
 - `protocol` (required): model API protocol, one of `"anthropic"`, `"openai_responses"`, `"openai_chat"`, or `"google"`
 - `reasoning_effort_levels` (optional): list of supported reasoning effort levels, for example `["low", "medium", "high"]`. The latest value in the list is used as the default
+- `interleaved_reasoning` (optional, default `false`): if thinking tokens are sent as a dedicated `reasoning_content` field (`true`) or inline in message text (`false`). Applies only when using the `openai_chat` protocol
 - `subscription` (optional): `"zen"`, `"go"`, or `"free"` (defaults to `"zen"`)
 - `custom_model_api_url` (optional): custom API base URL to use instead of the default OpenCode API