From 5573e0e8a2f9f3190c3afbf4632167f4173ba6bd Mon Sep 17 00:00:00 2001
From: Bryan Helmkamp <bryan@brynary.com>
Date: Mon, 23 Feb 2026 12:30:25 -0500
Subject: [PATCH] Fix truncated LLM responses by populating catalog max_output

Requests were always sent with max_tokens: None, which defaulted to
4096 in the Anthropic provider, causing large outputs to be truncated.
Now build_request() looks up the model's max_output from the catalog
and passes it through, giving each model its full output capacity.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 crates/agent/src/session.rs |  3 ++-
 crates/llm/src/catalog.json | 14 +++++++-------
 crates/llm/src/catalog.rs   |  1 +
 3 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/crates/agent/src/session.rs b/crates/agent/src/session.rs
index f9bd504b0..dc7794f4f 100644
--- a/crates/agent/src/session.rs
+++ b/crates/agent/src/session.rs
@@ -396,7 +396,8 @@ impl Session {
             response_format: None,
             temperature: None,
             top_p: None,
-            max_tokens: None,
+            max_tokens: llm::catalog::get_model_info(self.provider_profile.model())
+                .and_then(|m| m.max_output),
             stop_sequences: None,
             reasoning_effort: self.config.reasoning_effort.clone(),
             metadata: None,
diff --git a/crates/llm/src/catalog.json b/crates/llm/src/catalog.json
index 8f4c1e823..381f4a820 100644
--- a/crates/llm/src/catalog.json
+++ b/crates/llm/src/catalog.json
@@ -4,7 +4,7 @@
     "provider": "anthropic",
     "display_name": "Claude Opus 4.6",
     "context_window": 200000,
-    "max_output": null,
+    "max_output": 128000,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -17,7 +17,7 @@
     "provider": "anthropic",
     "display_name": "Claude Sonnet 4.5",
     "context_window": 200000,
-    "max_output": null,
+    "max_output": 64000,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -30,7 +30,7 @@
     "provider": "openai",
     "display_name": "GPT-5.2",
     "context_window": 1047576,
-    "max_output": null,
+    "max_output": 128000,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -43,7 +43,7 @@
     "provider": "openai",
     "display_name": "GPT-5 Mini",
     "context_window": 1047576,
-    "max_output": null,
+    "max_output": 128000,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -56,7 +56,7 @@
     "provider": "openai",
     "display_name": "GPT-5.2 Codex",
     "context_window": 1047576,
-    "max_output": null,
+    "max_output": 128000,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -69,7 +69,7 @@
     "provider": "gemini",
     "display_name": "Gemini 3 Pro (Preview)",
     "context_window": 1048576,
-    "max_output": null,
+    "max_output": 65536,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
@@ -82,7 +82,7 @@
     "provider": "gemini",
     "display_name": "Gemini 3 Flash (Preview)",
     "context_window": 1048576,
-    "max_output": null,
+    "max_output": 65536,
     "supports_tools": true,
     "supports_vision": true,
     "supports_reasoning": true,
diff --git a/crates/llm/src/catalog.rs b/crates/llm/src/catalog.rs
index c01b896c7..6be91d4c2 100644
--- a/crates/llm/src/catalog.rs
+++ b/crates/llm/src/catalog.rs
@@ -52,6 +52,7 @@ mod tests {
         assert!(info.supports_vision);
         assert!(info.supports_reasoning);
         assert_eq!(info.context_window, 200_000);
+        assert_eq!(info.max_output, Some(128_000));
     }
 
     #[test]