AI Desk · Model Identity

Compare

Pick two models. The bench shows whether their measured fingerprints agree, exactly where they part, and what their declared contracts share.

Verdict

Tokenizer fingerprint

Test string	A	B	Verdict

Values are the marginal prompt-token cost of each string with the model's fixed overhead subtracted. An x marks a row excluded as corrupt for that model.

API surface

The declared serving contract, field by field

Field	A	B	Verdict

Try a comparison

Each opens a worked example with its verdict

Ox Alpha and Z.ai: GLM 5.3the stealth model and its likely familyopen Microsoft: Phi 4 and IBM: Granite 4.1 8Btwo labs, one shared open vocabularyopen StepFun: Step 3.7 Flash and DeepSeek: DeepSeek V3tokenizer reuse across labsopen Z.ai: GLM 5.3 and OpenAI: GPT-5 Nanotwo unrelated families divergingopen

Or choose two models above: the address updates so any comparison can be linked and cited.