The video showcases the promising initial tests of Anthropic’s Claude Fable 5 model in agentic AI trading, achieving a 71% win rate and notable profits by dynamically analyzing market data, adapting strategies in real-time, and balancing risk with innovative tactics. The presenter highlights the model’s transparency, cost management through model switching, and plans for ongoing monitoring, encouraging others to explore its potential while being mindful of token consumption.
The video discusses the initial testing of Anthropic’s newly released Claude Fable 5 model applied to agentic AI trading. The presenter shares impressive results from running the model overnight on a 5-minute up-and-down market from Poly Market, achieving a 71% win rate and a profit of over $41 in about 10 hours, with a total daily gain of around $82. The model’s strategy selection and performance exceeded expectations, prompting a deeper dive into how the model analyzed the data and developed its trading approach.
The setup involved collecting around 24 hours of market data just before the model’s release, which was then fed into Claude Fable 5 to analyze and generate a trading strategy aimed at maximizing expected value. The model meticulously examined snapshots of the market data, testing various signals and strategies. Notably, it incorporated trading fees into its calculations, which is crucial for realistic profit estimation. The resulting strategy included specific buy signals based on a fair value formula, trade frequency limits, risk management rules, and stop-loss parameters.
One innovative aspect highlighted was the model’s ability to adapt and refine its strategy during live trading. After an initial three-hour trading period, the presenter set up a monitoring system using scheduled jobs that woke the AI every two hours to review trade performance and adjust the strategy if necessary. This dynamic adjustment led to improved performance, including the identification of a unique “deep long shots fading a jump” tactic, where the model occasionally took higher-risk trades with lower probabilities but higher payoffs, balancing conservative and aggressive trading styles.
The presenter also noted the model’s token consumption, which was quite high, prompting the use of model switching to manage costs effectively. They demonstrated how Claude Fable 5 could generate an interactive HTML report explaining the trading strategy, including the formula for buy signals and a live decision engine simulation. This transparency helped the presenter understand the model’s decision-making process and highlighted some early mistakes that the model corrected over time, tightening entry points and improving overall trade quality.
In conclusion, the initial tests of Claude Fable 5 for agentic AI trading showed promising results with strong profitability and innovative strategy development. The presenter plans to continue running and monitoring the bot, sharing updates on a Discord channel. They recommend others interested in AI trading to explore Fable 5 but to be cautious about token usage and consider model switching. Overall, the video conveys excitement about the potential of this advanced AI model in automated trading and encourages viewers to follow for future updates.