pip install torch: No matching distribution found (tested)

ERROR: Could not find a version that satisfies the requirement torch (from versions: none) means no torch wheel matches your interpreter or platform. torch 2.14.1 ships wheels for CPython 3.10-3.14 on glibc Linux (x86_64/aarch64), Apple Silicon and 64-bit Windows only, with no source distribution. Python 3.15, Alpine, Intel Macs and 32-bit Python all hit this error.

1. Reproduce

b1.sh asks pip to resolve torch (metadata only, nothing is downloaded). It ran in two official Docker images: python:3-slim (Python 3.15, glibc) and python:3-alpine (Python 3.14, musl):

python -c "import sys, platform; print('Python', sys.version.split()[0], platform.libc_ver()[0] or 'musl')"
pip install --dry-run --no-deps torch 2>&1 | grep -v '^\[notice\]'
echo "pip exit status: ${PIPESTATUS[0]}"

### python:3-slim
Python 3.15.1 glibc
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch
pip exit status: 1
### python:3-alpine
Python 3.14.7 musl
ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch
pip exit status: 1

Both fail identically with (from versions: none): pip found zero torch files it is allowed to install, for two different reasons.

2. Why: torch ships wheels only, for a fixed matrix

b2.py lists every file of the latest release from the PyPI JSON API:

# Which Python versions and platforms does the latest torch ship wheels for?
import json, urllib.request
d = json.load(urllib.request.urlopen("https://pypi.org/pypi/torch/json"))
print("torch", d["info"]["version"], "requires_python:", d["info"]["requires_python"])
print("sdist on PyPI:", any(f["packagetype"] == "sdist" for f in d["urls"]))
plats = {}
for f in d["urls"]:
    _, _, py, _, plat = f["filename"][:-4].split("-")
    plats.setdefault(plat, set()).add(py)
for plat, pys in sorted(plats.items()):
    print(f"{plat:28} {' '.join(sorted(pys))}")

torch 2.14.1 requires_python: >=3.10
sdist on PyPI: False
macosx_14_0_arm64            cp310 cp311 cp312 cp313 cp314
manylinux_2_28_aarch64       cp310 cp311 cp312 cp313 cp314
manylinux_2_28_x86_64        cp310 cp311 cp312 cp313 cp314
win_amd64                    cp310 cp311 cp312 cp313 cp314

There is no source distribution, so if no wheel matches, pip has nothing to fall back to. Read the table against your setup:

• Python 3.15 (or anything outside 3.10-3.14): no cp315 wheel. That is the python:3-slim failure above.
• Alpine / musl: only manylinux (glibc) wheels exist, no musllinux, so even Python 3.14 on Alpine fails.
• Intel Mac: the only macOS wheels are arm64.
• Old Linux: manylinux_2_28 needs glibc 2.28 or newer.
• 32-bit Python on Windows: only win_amd64.

Check yours with python -c "import sys, platform, struct; print(sys.version, platform.machine(), platform.libc_ver(), struct.calcsize('P')*8, 'bit')".

3. Fix: a supported Python on a supported platform

# Fix: an environment on a supported Python (3.10-3.14, glibc Linux)
uv venv -q -p 3.13 tenv
tenv/bin/python --version
uv pip install -p tenv/bin/python --dry-run --no-deps torch 2>&1 | grep -E 'Would install|torch=='
# CPU-only build from PyTorch's own index
uv pip install -p tenv/bin/python --dry-run --no-deps torch --index-url https://download.pytorch.org/whl/cpu 2>&1 | grep -E 'torch=='

Python 3.13.5
Would install 1 package
 + torch==2.14.1
 + torch==2.14.1+cpu

Same machine, same index: with a 3.13 environment the resolver picks torch 2.14.1, and PyTorch's own CPU index resolves 2.14.1+cpu. Without uv, create the venv with a supported interpreter (python3.13 -m venv .venv) and, in Docker, use a glibc image such as python:3.13-slim instead of Alpine or the floating python:3-slim tag. For CUDA builds, copy the exact --index-url from the selector on pytorch.org.

Related: Torch not compiled with CUDA enabled, pip install tensorflow: no matching distribution found, pip install failed, pip error: externally-managed-environment.

Written with AI assistance. Measured on 2026-10-24 with pip in the python:3-slim (Python 3.15.1) and python:3-alpine (Python 3.14.7) Docker images, uv with Python 3.13.5, and the live PyPI JSON API on Linux x86_64. The scripts ran verbatim and the output was pasted unedited (blank lines removed). PlainTables is our extension; PyTorch and pip are free.