diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index d19159cdee75..917e27c03758 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -81,6 +81,25 @@ function normalizeMessages(
       .filter((msg): msg is ModelMessage => msg !== undefined && msg.content !== "")
   }
 
+  // Strip reasoning parts that have no valid provider signature — they cannot be sent
+  // as thinking blocks to Anthropic without a signature and would cause a 400 error.
+  // This is a defence-in-depth guard; the primary prevention is in message-v2.ts where
+  // reasoning parts from a different model are skipped before reaching this point.
+  if (model.api.npm === "@ai-sdk/anthropic" || model.api.npm === "@ai-sdk/amazon-bedrock") {
+    msgs = msgs
+      .map((msg) => {
+        if (msg.role !== "assistant" || !Array.isArray(msg.content)) return msg
+        const filtered = msg.content.filter((part) => {
+          if ((part as any).type !== "reasoning") return true
+          const opts = (part as any).providerOptions?.anthropic
+          return opts?.signature != null || opts?.redactedData != null
+        })
+        if (filtered.length === 0) return undefined
+        return { ...msg, content: filtered }
+      })
+      .filter((msg): msg is ModelMessage => msg !== undefined)
+  }
+
   if (model.api.id.includes("claude")) {
     const scrub = (id: string) => id.replace(/[^a-zA-Z0-9_-]/g, "_")
     msgs = msgs.map((msg) => {
diff --git a/packages/opencode/src/session/llm.ts b/packages/opencode/src/session/llm.ts
index 3e96e51593d0..f88448e6a50d 100644
--- a/packages/opencode/src/session/llm.ts
+++ b/packages/opencode/src/session/llm.ts
@@ -193,7 +193,20 @@ const live: Layer.Layer<
         },
       )
 
-      const canTool = input.model.capabilities.toolcall
+      // DeepSeek R1 does not honour the `tools` parameter regardless of which provider
+      // hosts it — neither the direct api.deepseek.com endpoint (model id: deepseek-reasoner,
+      // providerID: deepseek) nor the AWS Bedrock endpoint (model id: deepseek.r1-v1:0,
+      // providerID: amazon-bedrock) support function calling, even though models.dev
+      // incorrectly reports tool_call: true for both. When tools are sent, R1 ignores
+      // the definitions and writes the invocation as markdown text instead.
+      // Detect R1 via the provider-agnostic `family` field ("deepseek-thinking") which
+      // is set for all DeepSeek reasoning variants, or fall back to matching the known
+      // model IDs directly for cases where family is absent.
+      const isDeepSeekR1 =
+        input.model.family === "deepseek-thinking" ||
+        (input.model.providerID === "deepseek" && input.model.api.id.toLowerCase().includes("reasoner")) ||
+        input.model.api.id.toLowerCase().includes("deepseek.r1")
+      const canTool = input.model.capabilities.toolcall && !isDeepSeekR1
       const tools = canTool ? resolveTools(input) : {}
 
       // LiteLLM and some Anthropic proxies require the tools parameter to be present
diff --git a/packages/opencode/src/session/message-v2.ts b/packages/opencode/src/session/message-v2.ts
index 20528763b8b1..b67d7e9beeb1 100644
--- a/packages/opencode/src/session/message-v2.ts
+++ b/packages/opencode/src/session/message-v2.ts
@@ -786,11 +786,15 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* (
               ...(differentModel ? {} : { callProviderMetadata: providerMeta(part.metadata) }),
             })
         }
-        if (part.type === "reasoning") {
+        // Reasoning parts carry a provider-specific signature that is cryptographically
+        // bound to the model that generated them. They cannot be replayed to a different
+        // model — doing so causes Anthropic to return "thinking.signature: Field required".
+        // Skip them entirely when the historical message came from a different model.
+        if (part.type === "reasoning" && !differentModel) {
           assistantMessage.parts.push({
             type: "reasoning",
             text: part.text,
-            ...(differentModel ? {} : { providerMetadata: part.metadata }),
+            providerMetadata: part.metadata,
           })
         }
       }
diff --git a/packages/opencode/src/session/processor.ts b/packages/opencode/src/session/processor.ts
index 21f9329c6fce..a3cd0ff89ee5 100644
--- a/packages/opencode/src/session/processor.ts
+++ b/packages/opencode/src/session/processor.ts
@@ -442,6 +442,13 @@ export const layer: Layer.Layer<
               },
               { text: ctx.currentText.text },
             )).text
+            // Some reasoning models (e.g. DeepSeek R1) emit chain-of-thought inside
+            // <think>…</think> tags in the text stream rather than as dedicated
+            // reasoning events. Strip those blocks so they don't appear as visible
+            // text in the UI — they are stored separately via reasoning-start/delta.
+            if (ctx.model.capabilities.reasoning) {
+              ctx.currentText.text = ctx.currentText.text.replace(/<think>[\s\S]*?<\/think>\s*/g, "").trimStart()
+            }
             {
               const end = Date.now()
               ctx.currentText.time = { start: ctx.currentText.time?.start ?? end, end }