Research: Anthropic Open-Sources FlashProve to Verify LLM Inference With Zero-Knowledge Proofs

November 3: Anthropic has released FlashProve, an open-source research prototype that uses zero-knowledge proofs to verify whether a large model's responses were generated by specified model weights, without needing to make those weights public. Anthropic says Claude participated in improving the system; in tests on the 400-billion-parameter Llama 4 Maverick model, a single H100 GPU generated proofs at about 110 tokens (text units) per second, but the required GPU time was roughly 70 to 95 times that of ordinary inference. The prototype currently supports contexts of up to 1,024 tokens and remains at the research stage.