Microsoft

Comparing 4 models from Microsoft.

Model Total ▼ Released Context SizeCreative writingRule followingUtilityMathematicsToolingLanguageLogic
Phi-3.5 Mini 128k55.49%Aug 17, 24128k3.8B18.74%44.81%61.44%80.00%55.64%59.00%88.75%
WizardLM 2 8x22b52.87%Apr 15, 2465k8x22B8.45%39.41%62.06%50.00%59.36%73.07%63.13%
Phi-3 Mini 128k49.87%Jul 1, 24128k3.8B21.40%35.96%54.89%85.00%42.44%58.12%94.38%
Phi-3 Medium 128k43.98%Apr 21, 24128k14B17.53%37.23%46.94%50.00%30.26%62.92%75.63%
Model Performance
Cost vs Performance

Compares total benchmark cost against overall score for Microsoft models. Quadrant lines are drawn at the median values.