diff --git a/candle-onnx/src/eval.rs b/candle-onnx/src/eval.rs
index 10a3b9377b..f7203b36f7 100644
--- a/candle-onnx/src/eval.rs
+++ b/candle-onnx/src/eval.rs
@@ -1274,6 +1274,30 @@ fn simple_eval_(
                 let output = candle_nn::ops::leaky_relu(input, alpha.into())?;
                 values.insert(node.output[0].clone(), output);
             }
+            // https://github.com/onnx/onnx/blob/main/docs/Operators.md#Gemm
+            "Gemm" => {
+                let a = get(&node.input[0])?;
+                let b = get(&node.input[1])?;
+                let c = get(&node.input[2])?;
+
+                let alpha = get_attr_opt::<f32>(node, "alpha")?.copied().unwrap_or(1.0);
+                let beta = get_attr_opt::<f32>(node, "beta")?.copied().unwrap_or(1.0);
+
+                let alpha = Tensor::full(alpha, a.shape(), &Device::Cpu)?;
+                let beta = Tensor::full(beta, c.shape(), &Device::Cpu)?;
+
+                let trans_a = get_attr_opt::<i64>(node, "transA")?.copied().unwrap_or(0);
+                let trans_b = get_attr_opt::<i64>(node, "transB")?.copied().unwrap_or(0);
+
+                let a = if trans_a == 0 { a.clone() } else { a.t()? };
+                let b = if trans_b == 0 { b.clone() } else { b.t()? };
+
+                let output = a
+                    .broadcast_mul(&alpha)?
+                    .broadcast_matmul(&b)?
+                    .broadcast_add(&c.broadcast_mul(&beta)?)?;
+                values.insert(node.output[0].clone(), output);
+            }
             op_type => bail!("unsupported op_type {op_type} for op {node:?}"),
         }
     }