diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
index 2a0d37e666e8..f5a8bb751000 100644
--- a/pkgs/by-name/ol/ollama/package.nix
+++ b/pkgs/by-name/ol/ollama/package.nix
@@ -94,7 +94,7 @@ let
 
   cudaToolkit = buildEnv {
     # ollama hardcodes the major version in the Makefile to support different variants.
-    # - https://github.com/ollama/ollama/blob/v0.20.3/CMakePresets.json#L21-L47
+    # - https://github.com/ollama/ollama/blob/v0.20.4/CMakePresets.json#L21-L47
     name = "cuda-merged-${cudaMajorVersion}";
     paths = map lib.getLib cudaLibs ++ [
       (lib.getOutput "static" cudaPackages.cuda_cudart)
@@ -140,13 +140,13 @@ let
 in
 goBuild (finalAttrs: {
   pname = "ollama";
-  version = "0.20.3";
+  version = "0.20.4";
 
   src = fetchFromGitHub {
     owner = "ollama";
     repo = "ollama";
     tag = "v${finalAttrs.version}";
-    hash = "sha256-o9iCqdOfNMxfIyThQAOSSQZE2ZyBuyJWFr6wqvQo1A0=";
+    hash = "sha256-8TbZvxxaUdROpe3gnBx0XzX62tbQ9QeJP3Yp7XXJoTQ=";
   };
 
   vendorHash = "sha256-Lc1Ktdqtv2VhJQssk8K1UOimeEjVNvDWePE9WkamCos=";
@@ -232,7 +232,7 @@ goBuild (finalAttrs: {
     '';
 
   # ollama looks for acceleration libs in ../lib/ollama/ (now also for CPU-only with arch specific optimizations)
-  # https://github.com/ollama/ollama/blob/v0.20.3/docs/development.md#library-detection
+  # https://github.com/ollama/ollama/blob/v0.20.4/docs/development.md#library-detection
   postInstall = ''
     mkdir -p $out/lib
     cp -r build/lib/ollama $out/lib/