Descriptions:
The All About AI channel puts OpenAI’s newly released GPT 5.6 Soul Max to work on a real-world quantitative trading task: analyzing 610,000 rows of Polymarket order book data spanning 43 BTC up/down markets to identify a profitable strategy. The host runs the same prompt in parallel on GPT 5.6 Soul Max and GPT 5.5 Extra High via Claude Codex, using a detailed slash-goal prompt specifying Monte Carlo simulation best practices, drawdown constraints, and fee avoidance — then compares how each model approaches the fair value strategy that has been producing results in prior sessions (including $213 profit in the prior week).
Before diving into the live test, the video provides brief benchmark context from Artificial Analysis, showing GPT 5.6 Soul scoring 59 on their intelligence index versus 5.5’s 55. The host encounters a new Codex safety-check prompt mid-run tied to Binance API calls, documenting the interruption on screen. The comparison focuses less on synthetic benchmarks and more on whether model behavior diverges in practice when given the same open-ended quant research goal.
For developers and traders exploring agentic coding tools for data-intensive tasks, the video offers an honest look at what it actually feels like to run GPT 5.6 on a non-trivial autonomous research loop, including usage consumption against a $200 Codex plan.
📺 Source: All About AI · Published July 09, 2026
🏷️ Format: Hands On Build
