On November 2, 2026, Anthropic published its FlashProve research and open-source prototype. By November 3, traders were discussing the announcement in connection with StarkWare and STRK. No FlashProve deployment on Starknet, use of STRK, or commercial partnership with the network was announced.
What FlashProve Can Actually Verify
Customers using an AI model through an API or routing service generally cannot verify that the model they paid for produced the response. A service could substitute a cheaper model, or an attacker could tamper with the output. FlashProve aims to make such substitutions detectable.
The system uses zkSNARKs, a type of cryptographic proof. A provider first publishes a commitment to a fixed set of model weights. It can then prove that a particular computation was performed using those weights, without exposing them to the verifier. The model's architecture is not hidden.
There are two limits worth separating. A valid proof does not establish that the model's answer is factually correct. And the published benchmark did not verify entire, arbitrarily long replies: it checked one selected next-token computation per prompt.
What 400 Billion Parameters and 110 Tokens per Second Mean
Anthropic tested FlashProve on Llama 4 Maverick, a mixture-of-experts model with approximately 400 billion parameters, about 17 billion of which are active for a given token. The benchmark used a 4-bit quantized, integer-based version of the model. Anthropic says Claude helped develop and optimize the proof system, but the benchmark did not measure Claude's response speed.
The test processed a batch of 24 requests, each with a 1,024-token context, using one NVIDIA H100 GPU for proof generation. The reported 110 tokens per second measures context tokens processed while generating proofs, not the rate at which an AI chatbot writes its answers. Each proof was about 19 MB, and verification took roughly 17 seconds on 24 CPU threads.
The trade-off remains substantial: proof generation consumed about 70–95 times more GPU time than comparable inference with vLLM. FlashProve is an open-source research prototype, not a production service; its repository explicitly states that it is not maintained.
The Actual Link Between Anthropic, StarkWare and Starknet
Alessandro Chiesa, listed among the research contributors, is affiliated with EPFL and StarkWare. That establishes a genuine research connection. It does not establish a commercial relationship between Anthropic and the Starknet network.
FlashProve uses a zkSNARK construction. Starknet relies on STARK proofs to verify blockchain computations. Both are approaches to verifiable computation, but they are not the same protocol. Nothing in the FlashProve release says that its proofs are submitted to Starknet or that verification requires STRK.
STRK serves existing functions in the Starknet ecosystem, including transaction fees, governance and staking. The FlashProve announcement does not add another use case for the token or show how the research would increase on-chain activity. Scientific relevance is not the same as token demand.
Why STRK Drew Traders' Attention
The narrative is easy to trade: Anthropic, Claude, zero-knowledge proofs and StarkWare all appear in one research announcement. Traders may speculate that verifiable AI will eventually benefit blockchain projects working in adjacent technology. That expectation can move a token before any commercial adoption exists.
On November 3, a trader reported that market monitoring had flagged a Bull Trap setup on STRK. The report is worth investigating, but an alert is not proof that a bull trap has played out. Without the recorded price chart, open interest and the exact alert conditions, neither the setup's outcome nor FlashProve's role in the price move can be established.
A weak fundamental link does not put a ceiling on a speculative rally. Shorting solely because the announcement contains no Starknet partnership can be costly.
What Would Confirm a Bull Trap?
A bull trap takes shape when a breakout attracts buyers but fails to hold, leaving late longs exposed to a reversal. Liquidations can accelerate the drop, particularly when leverage builds during the initial rally.
A stronger bearish case would include rising OI during the pump, a failed hold above the breakout level, fading buying pressure and a subsequent price decline. Falling OI after the reversal can indicate position unwinds, but it does not identify which side is closing without further evidence.
The setup also needs an invalidation condition. If STRK reclaims the breakout level, holds it and continues to attract buyers, the bull-trap thesis is not confirmed. Trying to call the exact top of a news-driven move creates significant risk for leveraged shorts.
Checking the Setup with Crypto Resources
Trap Radar PRO lets traders assess price action alongside open interest (OI), CVD, funding, volume and liquidations. Instead of treating one spike as a trade signal, a bull-trap analysis follows the sequence: positions build during the rally, buying momentum weakens, and price responds to a retest of the high.
The Pump, OI and liquidation screeners help locate unusual activity and determine whether a reversal coincides with position unwinds. Market Median adds context from the broader market using RegDev, Median RSI and the percentage of assets above SMA200. An isolated STRK pump during weak market breadth deserves a different assessment from a rally supported across the market.
These tools cannot establish why STRK moved or predict the exact top. They help verify conditions and define risk. The basic hourly pump and liquidation screeners on Crypto Resources are free without registration; Market Median is free after registration.
What Would Make FlashProve Relevant to STRK?
For FlashProve, the next milestones are lower proving costs, support for longer verified outputs and independent review of the implementation. Its current performance figures come from a specific experimental setup, not a general-purpose production rollout.
For Starknet, a confirmed integration or an application using its network would be a more meaningful development than shared expertise alone. For STRK, the missing step is an identifiable link between such adoption and network fees, staking or another source of token demand.
FlashProve is a meaningful cryptography release, but it is not yet a verified fundamental catalyst for STRK. For traders, price structure, OI and liquidity matter more than the association in the headline.
Frequently Asked Questions
Has Anthropic partnered with Starknet?
No such partnership has been announced. Alessandro Chiesa, who lists StarkWare among his affiliations, contributed to FlashProve, but that is not a Starknet integration or an STRK agreement.
Does FlashProve run on Starknet?
No integration has been announced. FlashProve is a research prototype for proving AI computations; Starknet uses STARK proofs for blockchain execution.
Does FlashProve make Claude generate 110 tokens per second?
No. The figure refers to context processing during proof generation in a benchmark using Llama 4 Maverick. The test verified one selected next-token computation per request.
Why did traders connect STRK to Anthropic's research?
Because a StarkWare-affiliated researcher contributed and both projects work in cryptographic verification. That connection does not demonstrate a commercial partnership or new demand for STRK.
Is a Bull Trap alert enough to short STRK?
No. The setup needs confirmation from price action, OI, volume and liquidations, plus a clear invalidation level. A news-driven rally can keep running despite an unconvincing fundamental narrative.
Risk Disclaimer
This article is for informational and analytical purposes only and does not constitute investment advice. Market reactions and interpretations of trading signals can change quickly. Leveraged cryptocurrency trading carries a substantial risk of loss.