Key Notes
- STRK rose 9.8% from its 11:00 UTC Coinbase open to an intraday peak of $0.04522.
- FlashProve processed about 110 context tokens per second in a batched Llama 4 Maverick benchmark.
- Anthropic credits a StarkWare researcher and the cause of the rally remains unconfirmed.
Starknet’s STRK token climbed from $0.04120 at 11:00 UTC on November 3 to $0.04522 in the 11:45–11:50 UTC window, a gain of about 9.8%. It later traded at $0.04350 at 12:33 UTC, about 5.6% above the 11:00 level. Coinbase market
The move came a day after Anthropic published FlashProve, an open-source prototype for verifying which AI model produced an output. The research credits Alessandro Chiesa of EPFL and StarkWare, giving the project a connection to the team behind Starknet. The market data leaves the cause of the rally unconfirmed. Anthropic research
What the 110 Tokens per Second Measure
Anthropic tested FlashProve on a 4-bit version of Meta’s Llama 4 Maverick, a mixture-of-experts model with 400 billion total parameters. A single Nvidia H100 processed roughly 110 context tokens per second across 24 batched requests. Each request used a 1,024-token window and the benchmark checked one output token per proof. Project repository
Anthropic says Claude helped optimize the proving system. Each proof was about 19 MB, and CPU verification took about 17 seconds using 24 threads. Producing the proofs required roughly 70 to 95 times the GPU time of ordinary inference in Anthropic’s comparison. Benchmark results
How FlashProve Checks a Model’s Output
A provider publishes a cryptographic commitment to a set of model weights. A verifier can then check that a tested output token matches what those weights produce for the supplied input, without seeing the weights. The prototype uses a zkSNARK and checks an exact integer version of the model under greedy decoding. Research details
Where StarkWare Fits
Anthropic cites Starknet as an example of a blockchain system using related proof techniques. Starknet documents STARK proofs and its S-two prover for validating network computation. Chiesa’s participation is a research link. Anthropic’s release announces neither a Starknet deployment nor a role for STRK. Starknet prover
Starknet’s own work on proof systems has also featured in its independent L1 research, which the community backed in an October advisory vote.
Disclaimer: CoinScreamer is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and market insights on digital assets and related technologies. NuvexMedia LLC invests in and collaborates with companies across the digital asset, blockchain, and technology sectors. These relationships do not influence CoinScreamer’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2025 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
