Prompt Reliability Through Intelligent Semantic Multiplexing
PRISM evaluates whether a language model maintains consistent, correct, and instruction-compliant behavior when the same underlying question is expressed through different prompt formulations.
Download the desktop application to run the benchmark locally, or explore previously evaluated models through the Results section.
How PRISM works
1
Evaluate
Run a standardized benchmark locally against supported language models.
2
Analyze
Examine accuracy, consistency, instruction compliance, and sensitivity to prompt formulation.
3
Compare
Compare locally generated results with previously evaluated models.
Run PRISM locally
Download the PRISM desktop application to evaluate supported local language models on your own hardware. Benchmark execution happens entirely on your machine.
Downloads aren't available right now. Check the Releases page on GitHub directly.