Grok 4.20 wins Alpha Arena Season 1.5 live trading contest

Alpha Arena runs frontier models as autonomous traders with real money and equal starting capital. Season 1.5 ended with Grok 4.20 in first place. According to Briefs, it grew $10,000 to about $11,060 and was reportedly the only entrant to close in the black, edging out OpenAI's model.

The format tests something static benchmarks miss: sequential decisions under uncertainty, with feedback from a market that does not reset. Each model receives market data, decides on positions and lives with the results. That makes it a rough proxy for agentic reliability in high-stakes, noisy environments.

The result needs heavy caveats:

- One season is a tiny sample, and returns of about 10% over a short window can reflect market direction or luck as much as skill.

- The other models losing money may say more about overtrading than about intelligence.

- Results depend on prompts, tool access and position limits, which vary by contest design.

For xAI, the win is a marketing point in a week when its main news was environmental conflict over generators at Colossus 2. Watch whether the result holds in later seasons. Also watch whether xAI connects it to a finance product, since Grok's integration with X gives it a natural distribution channel for market commentary.