Anthropic releases FlashProve, a prototype for cryptographically proving AI inference
Anthropic has open-sourced FlashProve, a zkSNARK-based system that lets users verify a response really came from the model a provider claims.
By GennnAi··1 min read
On Llama 4 Maverick (400 billion parameters, 4-bit quantization), a single H100 GPU generates proofs at about 110 tokens per second, with 19 MB per 1,024-token request. That costs 70 to 95 times the GPU time of running the model on vLLM. The proofs hide model weights, not the architecture. Anthropic released code, Lean proofs and documentation under Apache 2.0, presenting it as intermediate work.
Source: Anthropic, published on November 2, 2026.