Daily AI news, sourcedOn Nov 3, Anthropic open-sources FlashProve for zkSNARK provable inference on large MoE models.
Where it stands
On Nov 3, Anthropic open-sources FlashProve for zkSNARK provable inference on large MoE models.
- Official
Anthropic open-sources FlashProve for zkSNARK provable inference on large MoE models
Anthropic published this on Nov 3, 2026 (source: anthropic.com).
Anthropic’s Frontier Red Team published Progress on provable inference, releasing FlashProve, an open-source zkSNARK prototype that proves a committed MoE model produced a given response without revealing weights. On Llama 4 Maverick (400B MoE, 4-bit), a single H100 proves about 110 tokens/second with ~19 MB proofs per 1,024-token request (~70–95× vLLM GPU time). The release includes a technical report, protocol notes, prover/verifier code (C++ plus Lean), and Lean proofs under Apache 2.0; it is framed as intermediate research progress, not a finished product feature.
Source: Anthropic
- Official
- An institution announces something that has already happened, through its own channel or a founder or executive account.
- Statement by …
- A named person's views, predictions, plans or comments, including remarks reported by media when the original is unavailable.
- Reported by …
- Only media or another third party is saying it. We name the outlet.
- Unconfirmed
- Circulating online (posts, comments, leaks). No company, named person or outlet has confirmed it.
Items older than 72 hours never go in Top stories and show their original date.