Anthropic Says FlashProve Generates Llama 4 Maverick Proofs at 110 Tokens a Second
Summary
Anthropic releases FlashProve, an open-source prototype that generates cryptographic proofs that a specific model produced a given response. Anthropic says a single H100 proves Llama 4 Maverick at about 110 tokens per second, with 18.78 MB proofs verified in seconds, roughly 70 to 95 times slower than plain inference on vLLM. Contexts cap at 1,024 tokens.
Key Points
- Anthropic says the offline commitment phase runs once per model and takes about 39 minutes.
- Anthropic says the commitment stays valid for roughly 16 million requests before a provider must commit again.
- Anthropic reports proof verification takes 17 seconds on 24 threads and 21 seconds on one thread.