diff --git a/crates/arc-llm/src/catalog.json b/crates/arc-llm/src/catalog.json
index ef78dd387..b39422140 100644
--- a/crates/arc-llm/src/catalog.json
+++ b/crates/arc-llm/src/catalog.json
@@ -10,6 +10,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 15.0,
     "output_cost_per_million": 75.0,
+    "estimated_output_tps": 25,
     "aliases": ["opus", "claude-opus"]
   },
   {
@@ -23,6 +24,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 3.0,
     "output_cost_per_million": 15.0,
+    "estimated_output_tps": 50,
     "aliases": ["sonnet", "claude-sonnet"]
   },
   {
@@ -36,6 +38,7 @@
     "supports_reasoning": false,
     "input_cost_per_million": 0.8,
     "output_cost_per_million": 4.0,
+    "estimated_output_tps": 100,
     "aliases": ["haiku", "claude-haiku"]
   },
   {
@@ -49,6 +52,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 1.75,
     "output_cost_per_million": 14.0,
+    "estimated_output_tps": 65,
     "aliases": ["gpt5"]
   },
   {
@@ -62,6 +66,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 0.25,
     "output_cost_per_million": 2.0,
+    "estimated_output_tps": 70,
     "aliases": ["gpt5-mini"]
   },
   {
@@ -75,6 +80,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 1.75,
     "output_cost_per_million": 14.0,
+    "estimated_output_tps": 100,
     "aliases": []
   },
   {
@@ -88,6 +94,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 1.75,
     "output_cost_per_million": 14.0,
+    "estimated_output_tps": 100,
     "aliases": ["codex"]
   },
   {
@@ -101,6 +108,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 2.0,
     "output_cost_per_million": 12.0,
+    "estimated_output_tps": 85,
     "aliases": ["gemini-pro"]
   },
   {
@@ -114,6 +122,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 0.5,
     "output_cost_per_million": 3.0,
+    "estimated_output_tps": 150,
     "aliases": ["gemini-flash"]
   },
   {
@@ -127,6 +136,7 @@
     "supports_reasoning": false,
     "input_cost_per_million": 0.6,
     "output_cost_per_million": 3.0,
+    "estimated_output_tps": 50,
     "aliases": ["kimi"]
   },
   {
@@ -140,6 +150,7 @@
     "supports_reasoning": false,
     "input_cost_per_million": 0.6,
     "output_cost_per_million": 2.2,
+    "estimated_output_tps": 100,
     "aliases": ["glm", "glm4"]
   },
   {
@@ -153,6 +164,7 @@
     "supports_reasoning": false,
     "input_cost_per_million": 0.3,
     "output_cost_per_million": 1.2,
+    "estimated_output_tps": 45,
     "aliases": ["minimax"]
   },
   {
@@ -166,6 +178,7 @@
     "supports_reasoning": true,
     "input_cost_per_million": 0.25,
     "output_cost_per_million": 0.75,
+    "estimated_output_tps": 1000,
     "aliases": ["mercury"]
   }
 ]
diff --git a/crates/arc-llm/src/cli.rs b/crates/arc-llm/src/cli.rs
index 4a86c1978..100371627 100644
--- a/crates/arc-llm/src/cli.rs
+++ b/crates/arc-llm/src/cli.rs
@@ -89,21 +89,29 @@ fn format_cost(cost: Option<f64>) -> String {
     }
 }
 
+fn format_speed(tps: Option<f64>) -> String {
+    match tps {
+        None => "-".to_string(),
+        Some(t) => format!("{} tok/s", t as i64),
+    }
+}
+
 fn print_models_table(models: &[crate::types::ModelInfo]) {
     println!(
-        "{:<24} {:<12} {:<24} {:>10}  {:>7} {:>7}",
-        "MODEL", "PROVIDER", "ALIASES", "CONTEXT", "COST", ""
+        "{:<24} {:<12} {:<24} {:>10}  {:>7} {:>7}   {:>10}",
+        "MODEL", "PROVIDER", "ALIASES", "CONTEXT", "COST", "", "SPEED"
     );
     for model in models {
         let aliases = model.aliases.join(", ");
         println!(
-            "{:<24} {:<12} {:<24} {:>10}  {:>7} / {:<7}",
+            "{:<24} {:<12} {:<24} {:>10}  {:>7} / {:<7}  {:>10}",
             model.id,
             model.provider,
             aliases,
             format_context_window(model.context_window),
             format_cost(model.input_cost_per_million),
-            format_cost(model.output_cost_per_million)
+            format_cost(model.output_cost_per_million),
+            format_speed(model.estimated_output_tps)
         );
     }
 }
@@ -366,8 +374,8 @@ async fn test_models(provider: Option<&str>, model: Option<&str>) -> Result<()>
     }
 
     println!(
-        "{:<24} {:<12} {:>10}  {:>7}   {:>7}  {}",
-        "MODEL", "PROVIDER", "CONTEXT", "COST", "", "RESULT"
+        "{:<24} {:<12} {:>10}  {:>7}   {:>7}  {:>10}  {}",
+        "MODEL", "PROVIDER", "CONTEXT", "COST", "", "SPEED", "RESULT"
     );
 
     let mut failures = 0u32;
@@ -393,12 +401,13 @@ async fn test_models(provider: Option<&str>, model: Option<&str>) -> Result<()>
         };
 
         println!(
-            "{:<24} {:<12} {:>10}  {:>7} / {:<7}  {status}",
+            "{:<24} {:<12} {:>10}  {:>7} / {:<7}  {:>10}  {status}",
             info.id,
             info.provider,
             format_context_window(info.context_window),
             format_cost(info.input_cost_per_million),
-            format_cost(info.output_cost_per_million)
+            format_cost(info.output_cost_per_million),
+            format_speed(info.estimated_output_tps)
         );
     }
 
diff --git a/crates/arc-llm/src/types.rs b/crates/arc-llm/src/types.rs
index 639e9eff3..076becbcb 100644
--- a/crates/arc-llm/src/types.rs
+++ b/crates/arc-llm/src/types.rs
@@ -639,6 +639,7 @@ pub struct ModelInfo {
     pub supports_reasoning: bool,
     pub input_cost_per_million: Option<f64>,
     pub output_cost_per_million: Option<f64>,
+    pub estimated_output_tps: Option<f64>,
     pub aliases: Vec<String>,
 }