Anthropic releases FlashProve to verify which model produced a response

Anthropic introduced FlashProve, a prototype that lets a provider prove a response came from the model it committed to using without disclosing its weights. In a test with a 4-bit version of Llama 4 Maverick, a single H100 generated proofs at around 110 tokens per second. The proof for a 1,024-token request was about 19 MB. Anthropic estimates that generating the proof took roughly 70 to 95 times the GPU time required to run the model itself with vLLM.