diff --git a/pkgs/development/python-modules/flashinfer/default.nix b/pkgs/development/python-modules/flashinfer/default.nix
index 5c4c653ed981..905dc5d47a04 100644
--- a/pkgs/development/python-modules/flashinfer/default.nix
+++ b/pkgs/development/python-modules/flashinfer/default.nix
@@ -15,30 +15,21 @@
   ninja,
   numpy,
   torch,
+  pynvml,
+  einops,
 }:
 
-let
+buildPythonPackage rec {
   pname = "flashinfer";
-  version = "0.2.9";
-
-  src_cutlass = fetchFromGitHub {
-    owner = "NVIDIA";
-    repo = "cutlass";
-    # Using the revision obtained in submodule inside flashinfer's `3rdparty`.
-    tag = "v${version}";
-    hash = "sha256-d4czDoEv0Focf1bJHOVGX4BDS/h5O7RPoM/RrujhgFQ=";
-  };
-
-in
-buildPythonPackage {
-  format = "setuptools";
-  inherit pname version;
+  version = "0.2.14";
+  pyproject = true;
 
   src = fetchFromGitHub {
     owner = "flashinfer-ai";
     repo = "flashinfer";
     tag = "v${version}";
-    hash = "sha256-M0q6d+EpuTehbw68AQ73Fhwmw2tzjymYjSXaol9QC7Y=";
+    hash = "sha256-MZiZwdedz+Vxa1+VBfHDKf4NVSiOAytGboIJ0DvCXmk=";
+    fetchSubmodules = true;
   };
 
   build-system = [ setuptools ];
@@ -48,20 +39,16 @@ buildPythonPackage {
     ninja
     (lib.getBin cudaPackages.cuda_nvcc)
   ];
+
   dontUseCmakeConfigure = true;
 
-  buildInputs = [
-    cudaPackages.cuda_cudart
-    cudaPackages.libcublas
-    cudaPackages.cuda_cccl
-    cudaPackages.libcurand
+  buildInputs = with cudaPackages; [
+    cuda_cccl
+    cuda_cudart
+    libcublas
+    libcurand
   ];
 
-  postPatch = ''
-    rmdir 3rdparty/cutlass
-    ln -s ${src_cutlass} 3rdparty/cutlass
-  '';
-
   # FlashInfer offers two installation modes:
   #
   # JIT mode: CUDA kernels are compiled at runtime using PyTorch’s JIT, with
@@ -86,6 +73,8 @@ buildPythonPackage {
   dependencies = [
     numpy
     torch
+    pynvml
+    einops
   ];
 
   meta = with lib; {
diff --git a/pkgs/development/python-modules/kserve/default.nix b/pkgs/development/python-modules/kserve/default.nix
index 56c1382c95bd..10d86bf89974 100644
--- a/pkgs/development/python-modules/kserve/default.nix
+++ b/pkgs/development/python-modules/kserve/default.nix
@@ -158,6 +158,14 @@ buildPythonPackage rec {
   ];
 
   disabledTests = [
+    # AssertionError: assert CompletionReq...lm_xargs=None) == CompletionReq...lm_xargs=None)
+    "test_convert_params"
+
+    # Flaky: ray.exceptions.ActorDiedError: The actor died unexpectedly before finishing this task.
+    "test_explain"
+    "test_infer"
+    "test_predict"
+
     # Require network access
     "test_infer_graph_endpoint"
     "test_infer_path_based_routing"
diff --git a/pkgs/development/python-modules/mistral-common/default.nix b/pkgs/development/python-modules/mistral-common/default.nix
index e245f5b7e524..305581a4ef91 100644
--- a/pkgs/development/python-modules/mistral-common/default.nix
+++ b/pkgs/development/python-modules/mistral-common/default.nix
@@ -1,57 +1,96 @@
 {
   lib,
   buildPythonPackage,
-  fetchPypi,
-  poetry-core,
-  numpy,
-  pydantic,
+  fetchFromGitHub,
+
+  # build-system
+  setuptools,
+
+  # dependencies
   jsonschema,
+  numpy,
   opencv-python-headless,
-  sentencepiece,
-  typing-extensions,
-  tiktoken,
   pillow,
+  pydantic,
+  pydantic-extra-types,
   requests,
+  sentencepiece,
+  tiktoken,
+  typing-extensions,
+
+  # tests
+  click,
+  fastapi,
+  huggingface-hub,
+  openai,
+  pycountry,
+  pydantic-settings,
+  pytestCheckHook,
+  soundfile,
+  soxr,
+  uvicorn,
 }:
 
 buildPythonPackage rec {
   pname = "mistral-common";
-  version = "1.8.3";
+  version = "1.8.4";
   pyproject = true;
 
-  src = fetchPypi {
-    pname = "mistral_common";
-    inherit version;
-    hash = "sha256-DRl52CIntiX21xs8goF28FnajQ9aMwfN9TtIQJo5cKQ=";
+  src = fetchFromGitHub {
+    owner = "mistralai";
+    repo = "mistral-common";
+    tag = "v${version}";
+    hash = "sha256-HB6dsqiDSLhjyANk7ZT/cU98mjJamegAF0uKH8GfgM8=";
   };
 
-  pythonRelaxDeps = [
-    "pillow"
-    "tiktoken"
-  ];
-
-  build-system = [ poetry-core ];
+  build-system = [ setuptools ];
 
   dependencies = [
-    numpy
-    pydantic
     jsonschema
+    numpy
     opencv-python-headless
-    sentencepiece
-    typing-extensions
-    tiktoken
     pillow
+    pydantic
+    pydantic-extra-types
     requests
+    sentencepiece
+    tiktoken
+    typing-extensions
   ];
 
-  doCheck = true;
-
   pythonImportsCheck = [ "mistral_common" ];
 
-  meta = with lib; {
+  nativeCheckInputs = [
+    click
+    fastapi
+    huggingface-hub
+    openai
+    pycountry
+    pydantic-settings
+    pytestCheckHook
+    soundfile
+    soxr
+    uvicorn
+  ];
+
+  disabledTests = [
+    # Require internet
+    "test_download_gated_image"
+    "test_image_encoder_formats"
+    "test_image_processing"
+
+    # AssertionError: Regex pattern did not match.
+    "test_from_url"
+
+    # AssertionError, Extra items in the right set
+    "test_openai_chat_fields"
+  ];
+
+  meta = {
     description = "Tools to help you work with Mistral models";
     homepage = "https://github.com/mistralai/mistral-common";
-    license = licenses.asl20;
-    maintainers = with maintainers; [ bgamari ];
+    changelog = "https://github.com/mistralai/mistral-common/releases/tag/v${version}";
+    license = lib.licenses.asl20;
+    maintainers = with lib.maintainers; [ bgamari ];
   };
 }
diff --git a/pkgs/development/python-modules/openai-harmony/default.nix b/pkgs/development/python-modules/openai-harmony/default.nix
new file mode 100644
index 000000000000..74251b28ad4e
--- /dev/null
+++ b/pkgs/development/python-modules/openai-harmony/default.nix
@@ -0,0 +1,61 @@
+{
+  lib,
+  buildPythonPackage,
+  fetchFromGitHub,
+
+  # build-system
+  rustPlatform,
+
+  # dependencies
+  pydantic,
+
+  # optional-dependencies
+  fastapi,
+  uvicorn,
+}:
+
+buildPythonPackage rec {
+  pname = "openai-harmony";
+  version = "0.0.4";
+  pyproject = true;
+
+  src = fetchFromGitHub {
+    owner = "openai";
+    repo = "harmony";
+    rev = "v${version}";
+    hash = "sha256-2LOrMLrNR1D3isbjiv5w+1Ni9IMwMEPPTOnG1rm//ag=";
+  };
+
+  cargoDeps = rustPlatform.fetchCargoVendor {
+    inherit pname version src;
+    hash = "sha256-tQq6PFMYghIfJu8CddbXFKXxr41GJaElbCCQuSpnaqk=";
+  };
+
+  nativeBuildInputs = [
+    rustPlatform.cargoSetupHook
+    rustPlatform.maturinBuildHook
+  ];
+
+  dependencies = [
+    pydantic
+  ];
+
+  optional-dependencies = {
+    demo = [
+      fastapi
+      uvicorn
+    ];
+  };
+
+  pythonImportsCheck = [ "openai_harmony" ];
+
+  # Tests require internet access
+  doCheck = false;
+
+  meta = {
+    description = "Renderer for the harmony response format to be used with gpt-oss";
+    homepage = "https://github.com/openai/harmony";
+    license = lib.licenses.asl20;
+    maintainers = with lib.maintainers; [ GaetanLepage ];
+  };
+}
diff --git a/pkgs/development/python-modules/vllm/0004-drop-lsmod.patch b/pkgs/development/python-modules/vllm/0004-drop-lsmod.patch
deleted file mode 100644
index 3c80f8e24e65..000000000000
--- a/pkgs/development/python-modules/vllm/0004-drop-lsmod.patch
+++ /dev/null
@@ -1,18 +0,0 @@
---- a/setup.py
-+++ b/setup.py
-@@ -340,14 +340,7 @@ def _is_hpu() -> bool:
-         out = subprocess.run(["hl-smi"], capture_output=True, check=True)
-         is_hpu_available = out.returncode == 0
-     except (FileNotFoundError, PermissionError, subprocess.CalledProcessError):
--        if sys.platform.startswith("linux"):
--            try:
--                output = subprocess.check_output(
--                    'lsmod | grep habanalabs | wc -l', shell=True)
--                is_hpu_available = int(output) > 0
--            except (ValueError, FileNotFoundError, PermissionError,
--                    subprocess.CalledProcessError):
--                pass
-+        is_hpu_available = False
-     return is_hpu_available
- 
- 
diff --git a/pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch b/pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch
index 9bf7d3cdf4d8..f16982ece4a7 100644
--- a/pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch
+++ b/pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch
@@ -1,25 +1,12 @@
-From 7511784ceb9252091a9d63ac6b54dcc67dd2b262 Mon Sep 17 00:00:00 2001
-From: Conroy Cheers <conroy@corncheese.org>
-Date: Fri, 13 Jun 2025 17:42:10 +1000
-Subject: [PATCH] drop intel reqs
-
----
- requirements/cpu.txt | 3 ---
- 1 file changed, 3 deletions(-)
-
 diff --git a/requirements/cpu.txt b/requirements/cpu.txt
-index d7b0fc6d8..be2df751b 100644
+index d80354342..7434f32f0 100644
 --- a/requirements/cpu.txt
 +++ b/requirements/cpu.txt
-@@ -24,8 +24,5 @@ datasets # for benchmark scripts
- # cpu cannot use triton 3.3.0
- triton==3.2.0; platform_machine == "x86_64"
+@@ -21,7 +21,4 @@ torchvision; platform_machine != "ppc64le" and platform_machine != "s390x"
+ torchvision==0.22.0; platform_machine == "ppc64le"
+ datasets # for benchmark scripts
  
 -# Intel Extension for PyTorch, only for x86_64 CPUs
 -intel-openmp==2024.2.1; platform_machine == "x86_64"
--intel_extension_for_pytorch==2.7.0; platform_machine == "x86_64"
- py-libnuma; platform_system != "Darwin"
- psutil; platform_system != "Darwin"
--- 
-2.49.0
-
+-intel_extension_for_pytorch==2.6.0; platform_machine == "x86_64" # torch>2.6.0+cpu has performance regression on x86 platform, see https://github.com/pytorch/pytorch/pull/151218
+ triton==3.2.0; platform_machine == "x86_64" # Triton is required for torch 2.6+cpu, as it is imported in torch.compile.
diff --git a/pkgs/development/python-modules/vllm/default.nix b/pkgs/development/python-modules/vllm/default.nix
index cec9c00eadcc..b5e4edd75edc 100644
--- a/pkgs/development/python-modules/vllm/default.nix
+++ b/pkgs/development/python-modules/vllm/default.nix
@@ -5,7 +5,6 @@
   buildPythonPackage,
   pythonAtLeast,
   fetchFromGitHub,
-  fetchpatch,
   symlinkJoin,
   autoAddDriverRunpath,
 
@@ -45,6 +44,8 @@
   lm-format-enforcer,
   prometheus-fastapi-instrumentator,
   cupy,
+  cbor2,
+  pybase64,
   gguf,
   einops,
   importlib-metadata,
@@ -69,6 +70,8 @@
   bitsandbytes,
   flashinfer,
   py-libnuma,
+  setproctitle,
+  openai-harmony,
 
   # internal dependency - for overriding in overlays
   vllm-flash-attn ? null,
@@ -92,13 +95,13 @@ let
   shouldUsePkg =
     pkg: if pkg != null && lib.meta.availableOn stdenv.hostPlatform pkg then pkg else null;
 
-  # see CMakeLists.txt, grepping for GIT_TAG near cutlass
+  # see CMakeLists.txt, grepping for CUTLASS_REVISION
   # https://github.com/vllm-project/vllm/blob/v${version}/CMakeLists.txt
   cutlass = fetchFromGitHub {
     owner = "NVIDIA";
     repo = "cutlass";
-    tag = "v3.9.2";
-    hash = "sha256-teziPNA9csYvhkG5t2ht8W8x5+1YGGbHm8VKx4JoxgI=";
+    tag = "v4.0.0";
+    hash = "sha256-HJY+Go1viPkSVZPEs/NyMtYJzas4mMLiIZF3kNX+WgA=";
   };
 
   flashmla = stdenv.mkDerivation {
@@ -111,8 +114,8 @@ let
     src = fetchFromGitHub {
       owner = "vllm-project";
       repo = "FlashMLA";
-      rev = "575f7724b9762f265bbee5889df9c7d630801845";
-      hash = "sha256-8WrKMl0olr0nYV4FRJfwSaJ0F5gWQpssoFMjr9tbHBk=";
+      rev = "0e43e774597682284358ff2c54530757b654b8d1";
+      hash = "sha256-wxL/jtq/lsLg1o+4392KNgfw5TYlW6lqEVbmR3Jl4/Q=";
     };
 
     dontConfigure = true;
@@ -138,8 +141,8 @@ let
     src = fetchFromGitHub {
       owner = "vllm-project";
       repo = "flash-attention";
-      rev = "8798f27777fb57f447070301bf33a9f9c607f491";
-      hash = "sha256-UTUvATGN1NU/Bc8qo078q6bEgILLmlrjL7Yk2iAJhg4=";
+      rev = "57b4e68b9f9d94750b46de8f8dbd2bfcc86edd4f";
+      hash = "sha256-c7L7WZVVEnXMOTPBoSp7jhkl9d4TA4sj11QvOSWTDIE=";
     };
 
     dontConfigure = true;
@@ -161,7 +164,7 @@ let
 
   cpuSupport = !cudaSupport && !rocmSupport;
 
-  # https://github.com/pytorch/pytorch/blob/v2.7.0/torch/utils/cpp_extension.py#L2343-L2345
+  # https://github.com/pytorch/pytorch/blob/v2.7.1/torch/utils/cpp_extension.py#L2343-L2345
   supportedTorchCudaCapabilities =
     let
       real = [
@@ -231,6 +234,7 @@ let
     libcusolver # cusolverDn.h
     cuda_nvtx
     cuda_nvrtc
+    # cusparselt # cusparseLt.h
     libcublas
   ];
 
@@ -247,7 +251,7 @@ in
 
 buildPythonPackage rec {
   pname = "vllm";
-  version = "0.9.1";
+  version = "0.10.1.1";
   pyproject = true;
 
   # https://github.com/vllm-project/vllm/issues/12083
@@ -259,23 +263,12 @@ buildPythonPackage rec {
     owner = "vllm-project";
     repo = "vllm";
     tag = "v${version}";
-    hash = "sha256-sp7rDpewTPXTVRBJHJMj+8pJDS6wAu0/OTJZwbPPqKc=";
+    hash = "sha256-lLNjBv5baER0AArX3IV4HWjDZ2jTGXyGIvnHupR8MGM=";
   };
 
   patches = [
-    (fetchpatch {
-      name = "remove-unused-opentelemetry-semantic-conventions-ai-dep.patch";
-      url = "https://github.com/vllm-project/vllm/commit/6a5d7e45f52c3a13de43b8b4fa9033e3b342ebd2.patch";
-      hash = "sha256-KYthqu+6XwsYYd80PtfrMMjuRV9+ionccr7EbjE4jJE=";
-    })
-    (fetchpatch {
-      name = "fall-back-to-gloo-when-nccl-unavailable.patch";
-      url = "https://github.com/vllm-project/vllm/commit/aa131a94410683b0a02e74fed2ce95e6c2b6b030.patch";
-      hash = "sha256-jNlQZQ8xiW85JWyBjsPZ6FoRQsiG1J8bwzmQjnaWFBg=";
-    })
     ./0002-setup.py-nix-support-respect-cmakeFlags.patch
     ./0003-propagate-pythonpath.patch
-    ./0004-drop-lsmod.patch
     ./0005-drop-intel-reqs.patch
   ];
 
@@ -354,6 +347,7 @@ buildPythonPackage rec {
     aioprometheus
     blake3
     cachetools
+    cbor2
     depyf
     fastapi
     llguidance
@@ -366,6 +360,7 @@ buildPythonPackage rec {
     prometheus-fastapi-instrumentator
     py-cpuinfo
     pyarrow
+    pybase64
     pydantic
     python-json-logger
     python-multipart
@@ -393,6 +388,8 @@ buildPythonPackage rec {
     opentelemetry-api
     opentelemetry-exporter-otlp
     bitsandbytes
+    setproctitle
+    openai-harmony
     # vLLM needs Torch's compiler to be present in order to use torch.compile
     torch.stdenv.cc
   ]
@@ -476,8 +473,5 @@ buildPythonPackage rec {
       # find_isa
       "x86_64-darwin"
     ];
-    # ValueError: 'aimv2' is already used by a Transformers config, pick another name.
-    # Version bump ongoing in https://github.com/NixOS/nixpkgs/pull/429117
-    broken = true;
   };
 }
diff --git a/pkgs/top-level/python-packages.nix b/pkgs/top-level/python-packages.nix
index 1f45c475922e..365d496bd202 100644
--- a/pkgs/top-level/python-packages.nix
+++ b/pkgs/top-level/python-packages.nix
@@ -10797,6 +10797,8 @@ self: super: with self; {
 
   openai-agents = callPackage ../development/python-modules/openai-agents { };
 
+  openai-harmony = callPackage ../development/python-modules/openai-harmony { };
+
   openai-whisper = callPackage ../development/python-modules/openai-whisper { };
 
   openaiauth = callPackage ../development/python-modules/openaiauth { };