ConfidentApiError: Invalid API key after successful deepeval login using AnthropicModel (Python 3.13, DeepEval 4.0.9)

[Maheep25]

Environment

• OS: macOS
• Python: 3.13.9
• DeepEval: 4.0.9
• Judge Model: AnthropicModel("claude-sonnet-4-6")
• Anthropic SDK: 0.116.0

Description

I'm using AnthropicModel as the evaluation model (no OpenAI).

The evaluation itself completes successfully and the metric passes. The failure occurs only during the post-test upload to Confident AI (wrap_up_test_run() / post_test_run()).

The metric passes, then the following exception is raised:

deepeval.confident.types.ConfidentApiError: Invalid API key

What I verified

• deepeval login --api-key ... succeeds.
• CONFIDENT_API_KEY is loaded correctly.
• ANTHROPIC_API_KEY is loaded correctly.
• AnthropicModel is used successfully.
• The evaluation passes.
• curl https://api.confident-ai.com works.
• requests.get("https://api.confident-ai.com") works.

The issue only occurs during the upload step.

Console output

You're running DeepEval's latest Task Completion Metric!
(using claude-sonnet-4-6 (Anthropic))

est_case_0 (Passed 1 metrics)

ConfidentApiError: Invalid API key

I also intermittently see:

SSLCertVerificationError:
certificate verify failed:
unable to get local issuer certificate

although Python requests and curl both connect successfully to the same endpoint.

Code

from deepeval.models import AnthropicModel
from deepeval.metrics import TaskCompletionMetric

judge = AnthropicModel(
    model="claude-sonnet-4-6"
)

evaluate(
    test_cases=[test_case],
    metrics=[
        TaskCompletionMetric(
            threshold=0.7,
            model=judge
        )
    ]
)

Expected behavior

The evaluation should upload successfully after deepeval login.

Actual behavior

Evaluation succeeds locally, but upload fails with:

ConfidentApiError: Invalid API key

deepeval --version -> 4.0.9

python --version
Python 3.13.9

deepeval login
Successfully logged in

evals/test_TaskCompletion.py:22
evaluate(...)

deepeval.evaluate.evaluate.py
↓

global_test_run_manager.wrap_up_test_run()

↓

post_test_run()

↓

ConfidentApiError: Invalid API key

Additional debugging performed

I verified the following:

• deepeval login --api-key ... succeeds.
• CONFIDENT_API_KEY is loaded correctly from both .env and .env.local.
• AnthropicModel is used successfully.
• curl https://api.confident-ai.com returns HTTP 404 (expected).
• requests.get("https://api.confident-ai.com") returns HTTP 404.
• Python can verify SSL certificates using requests.
• The evaluation passes before the upload attempt.

The issue reproduces even after:

• deepeval login succeeds
• restarting the terminal
• restarting pytest

I also occasionally see:

SSLCertVerificationError:
certificate verify failed

However, both curl and Python requests connect successfully to
https://api.confident-ai.com, so I'm not sure whether this is related or simply a secondary warning.

Question:

Is DeepEval 4.0.9 officially supported on Python 3.13,
or is Python 3.12 currently the recommended runtime?

[penguine-ip]

Hey @Maheep25 thanks for the issue. Do you mind running deepeval diagnose and pasting the output?

This is must likely either an:

• API key mismatch error, or
• Region mismatch error

For the former, make sure you've copied the project API key.

[Maheep25]

Hey @Maheep25 thanks for the issue. Do you mind running deepeval diagnose and pasting the output?

This is must likely either an:

• API key mismatch error, or
• Region mismatch error

For the former, make sure you've copied the project API key.

Hi @penguine-ip, thanks for the quick response!

I ran deepeval diagnose and here are the relevant results:

DeepEval 4.0.9
Python 3.13.9

Confident AI

Status: Logged in
API key: ********I9yq0=
Data region: US
API endpoint: https://api.confident-ai.com

This indicates that:

deepeval login --api-key ... succeeds.
deepeval diagnose reports Status: Logged in.
The CONFIDENT_API_KEY is being picked up correctly from .env.local.
The inferred data region is US.

I also verified the following:

ANTHROPIC_API_KEY is loaded correctly.
CONFIDENT_API_KEY is loaded correctly.
AnthropicModel(model="claude-sonnet-4-6") is being used successfully.
The evaluation itself completes successfully and the metric passes.
curl https://api.confident-ai.com returns HTTP/2 404 (endpoint is reachable).
requests.get("https://api.confident-ai.com") also returns 404, so Python networking/SSL appears to work outside of DeepEval.

The only failure occurs after the evaluation finishes, during the upload step.

Console output:

✨ You're running DeepEval's latest Task Completion Metric!
(using claude-sonnet-4-6 (Anthropic))

✅ test_case_0 (Passed 1 metrics)

Immediately afterwards I see:

[Confident AI Trace Log]
Error posting trace:
SSLCertVerificationError:
certificate verify failed: unable to get local issuer certificate

and then:

deepeval.confident.types.ConfidentApiError: Invalid API key

One clarification regarding the API key:

Initially I was using an Organization API Key. After your suggestion, I created a new project (TestAgents) and checked the project settings. At the moment, I don't see a separate Project API Key—only the organization-level API keys are available in the UI.

Could you please confirm:

Does DeepEval require a Project API Key, or is the Organization API Key the correct one?
Could the SSL trace upload failure and the subsequent ConfidentApiError: Invalid API key be related, or are these two independent issues?

Happy to provide any additional logs or test a patch if needed.

[Eshaj20]

Hi, I'd like to investigate this — I suspect it may be a region-routing issue in _infer_region_from_api_key (api.py). Can I take this?