Anthropic releases open-source system for proving model inference

Anthropic has released FlashProve, an open-source system that generates cryptographic proofs that a response came from a specific model.

On a single H100 GPU, it proves inference for the 400-billion-parameter Llama 4 Maverick at over 100 tokens per second, with a roughly 19 MB proof per 1,024-token request.

The release includes code and Lean formal proofs under the Apache 2.0 license.