Claude Fable 5 flags crypto signals, misses timing and size

A desk test run in June evaluated Anthropic’s Claude Fable 5 on bitcoin, ether and XRP. The model was instructed to provide one key metric per coin, a price floor, a year-end range and a trigger that would invalidate each view. Testers compared the model’s claims with live on-chain data, ETF flow figures, stablecoin supply and prediction market odds.

For bitcoin, the model highlighted long-term holder behavior and said selling stopped in November 2025 and buying resumed afterward. It flagged ETF outflows near $401 million in May, set a floor at $52,000 to $56,000 and a year-end range of $78,000 to $92,000. On-chain data showed long-term holders turned net buyers in March, four months later than the model’s timing. ETF flow records for May show $2.43 billion in outflows and June already registered $1.81 billion in outflows. Prediction markets price year-end bitcoin between $60,000 and $65,000.

For ether, the model pointed to the validator entry queue as the main constraint on supply, and set a floor at $1,250 to $1,400 with a $2,000 to $2,600 year-end range. Validator queue data indicated about 3.03 million ETH in line on June 8, above the model’s threshold. The model’s suggested corporate forced-sale trigger did not materialize; reported large holders maintained or increased balances. ETF flow data show five straight months of outflows through March and an additional $540.88 million outflow in May. Prediction markets assign about 65% odds to ether trading below $1,250 and roughly 16% odds above $3,500.

For XRP, the model framed price action as a contest between monthly escrow releases and ETF buying. It set a $0.95 to $1.10 floor and a $2.20 to $2.60 top. ETF flows were mostly positive: seven inflow months out of eight since November, including $131.94 million in May and $10.06 million in June. Actual net releases from escrow measured near 128 million tokens, below the model’s assumed 300 million monthly leak. Prediction markets assign about an 80% chance of XRP trading under $1.00 and about 11% for levels above $2.60.

The test applied a single tiebreaker across calls: total stablecoin market capitalization. The model treated a float above $330 billion by Q4 as supporting bull ranges and below $280 billion as favoring bear floors. Stablecoin supply was $315.97 billion on June 10, down from a peak near $323 billion and $3.25 billion lighter over the prior week; neither threshold had been crossed.

Testers reported that the model consistently named relevant on-chain and flow indicators for each coin and provided specific numeric forecasts. In several cases the model’s timing and the magnitude of its flow estimates differed from the recorded data and market odds. The test compared each model claim against contemporaneous on-chain measures, ETF flow records, stablecoin supply figures and prediction market prices to document those differences.