What causes "curl_cffi is not available on this platform" in xtekky/gpt4free?

pip install -U curl_cffi to get a wheel matching your platform.

xtekky/gpt4free · error · ImportError

curl_cffi is not available on this platform

Error message

curl_cffi is not available on this platform

What it means

Defined in g4f/requests/curl_cffi.py as the constructor of a stub AsyncSession class that exists only when `from curl_cffi.requests import AsyncSession` failed at import time. Instantiating it raises ImportError('curl_cffi is not available on this platform'). The stub pattern lets the module import cleanly everywhere while making any actual use fail loudly. Common root cause is curl_cffi's compiled wheel being incompatible (e.g. illegal instruction on old CPUs), not just a missing package.

Source

Thrown at g4f/requests/curl_cffi.py:11

from __future__ import annotations

try:
    from curl_cffi.requests import AsyncSession, Response

    has_curl_cffi = True
except ImportError:
    # Fallback for systems where curl_cffi is not available or causes illegal instruction errors
    class AsyncSession:
        def __init__(self, *args, **kwargs):
            raise ImportError("curl_cffi is not available on this platform")

    class Response:
        pass

    has_curl_cffi = False

if has_curl_cffi:
    try:
        from curl_cffi import CurlMime

        has_curl_mime = True
    except ImportError:
        has_curl_mime = False
    try:
        from curl_cffi import CurlWsFlag

        has_curl_ws = True
    except ImportError:

View on GitHub (pinned to 973504e177)

Solutions

pip install -U curl_cffi to get a wheel matching your platform.
If it still fails, test `python -c "import curl_cffi"` — an illegal instruction or ImportError confirms a binary mismatch; try a different version (pip install 'curl_cffi==<older>') or a manylinux-compatible base image.
Switch the g4f backend to aiohttp (uninstall/avoid curl_cffi path) which needs no compiled curl.
On musl/Alpine, prefer a glibc-based image (python:3.x-slim) for wheel compatibility.

Defensive patterns

Strategy: validation

Validate before calling

from g4f.requests.curl_cffi import has_curl_cffi

if not has_curl_cffi:
    # use aiohttp backend instead, or fail fast with install instructions
    raise SystemExit("pip install -U curl_cffi")

Try / catch

try:
    s = AsyncSession()
except ImportError as e:
    if "curl_cffi" in str(e):
        from g4f.requests.aiohttp import StreamSession as FallbackSession  # non-impersonated fallback
        s = FallbackSession()

Prevention

Feature-detect has_curl_cffi at startup instead of waiting for a constructor failure.
Verify `import curl_cffi` in CI for every target platform/image.
On musl or old-CPU hosts, prefer glibc images or pin a curl_cffi version with compatible binaries.

When it happens

Trigger: Importing symbols from g4f.requests.curl_cffi and constructing AsyncSession on a machine where curl_cffi is uninstalled or its native wheel fails to load (unsupported CPU instructions, wrong manylinux level, broken install).

Common situations: Old CPUs lacking AVX/required instructions hitting curl_cffi's illegal-instruction issue; Alpine/musl images without a compatible wheel; pip resolving an old curl_cffi built for a different ABI; the comment in the source explicitly calls out the illegal-instruction fallback.

Related errors

AI-assisted analysis of xtekky/gpt4free@973504e177 (2026-08-14). Data as JSON: /api/errors/3f3c0252cdb4c279. Report an issue: GitHub.