APICHECK / AI TOOLBOX
AI Capability Benchmark: Test Model Degradation Against a Baseline
Evaluate arithmetic, logic, state tracking, code, context and instruction following. Inspect answers and scoring evidence and compare compatible baseline reports.
API connection
Open to guestsYour key stays in page memory and is forwarded through our server for requests. It is not saved to a database or browser storage. Refreshing clears it.
Capability check
Check six capability dimensions with objectively scored tasks. Repeat identical questions and compare a historical baseline for evidence of decline.
Run settings
Up to 24 calls. Output length follows the model provider defaults. Calls use your API credit. Stop at any time.
Enter your API URL, key, and model above to run this tool.
Results
Ready when you areYour next experiment starts here
Set up your API and run settings, then use the button on the left. Progress and results will appear here.
Compare a historical baseline
Save a run from a trusted state, repeat the same settings, and import the old report. Files stay on your device. Imported observations are user supplied, not provider-certified evidence.
No comparable baseline: this run describes current performance and cannot establish a decline.
How to use and interpret results
- Connect an API and choose a model and test mode.
- Run the suite and inspect questions, answers and scoring evidence.
- Compare against a baseline with the same suite, seed and settings; repeat anomalies.
Low scores alone do not prove model substitution or degradation. Sampling, prompts, provider settings, timeouts and test coverage affect results. A comparable baseline is needed to assess change.

