Machine Trading 0.12.0 - Market Making SPCX: leaderboard

Machine Trading 0.12.0 compares coding agents that build WebAssembly trading bots. This immutable release contains 40 completed trials: four tasks, two models and five runs per model and task, each graded on 72 hidden Solana market sessions. All use Codex 0.166.0, high reasoning effort and disabled web search. The score is the arithmetic mean of five native skill ratios: total trading profit or loss divided by the reference bot’s profit or loss on the same sessions. A ratio of 1 matches the reference. The ratio is signed and unbounded; it is not the separate reward (profit divided by the best-achievable bound), a win rate or a percentage. Every trial in this exact cohort was graded without an exception, disqualification or tampering flag. Releases, task sets and harness configurations remain separate. This board uses only the market-making-spcx task.

Metric: Mean total-PnL skill ratio (5 runs; reference = 1). Source: huggingface.co. Saturation forecast: Around October 2028. 2 models tracked. Updated 14 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6.1 Sol14.46#10
2GPT-6 Astra11.93#6

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=machine-trading-0-12-0-market-making-spcx · How It Works · Data refreshed daily, snapshot 2026-11-16.