The Chinese artificial intelligence model DeepSeek V3.1 has doubled the deposit in the crypto market in just nine days in the Alpha Arena benchmark, starting with $10,000 and reaching $23,000 on October 27. This is the best result among AI participants and yet another loud signal about the potential of algorithmic trading.
Who else is in the AI game and what are the results
The test is ongoing in real time with real funds, obligatory stop-losses, fixed take-profits, and the possibility of using leverage up to x20. After DeepSeek, Qwen3 follows closely, which also shows aggressive dynamics but lags behind the leader.
Model results as of October 27
-
DeepSeek V3.1 — $22,200
-
Qwen3 — ~$18,400
-
Claude Sonnet 4.5 — $12,200
-
Grok 4 — ~$11,000
-
Gemini 2.5 Pro and GPT-5 — $4,000 each (approximately 60% losses).
Experts note different trading styles among AI. DeepSeek operates disciplined, makes few transactions, and keeps part of the capital in reserve. Qwen3 aggressively longs Bitcoin with 20x leverage. Grok 4 bets on Dogecoin. Claude Sonnet 4.5 trades conservatively and quickly locks in profits. The most unsuccessful strategies are in GPT-5 and Gemini — they tried to catch reversals, but the market proved stronger.
The first season of Alpha Arena will end on November 3, after which a new one with updated rules will launch. The organizers say that the goal is to test AI in the most chaotic environment, which is the crypto market.
Meanwhile, Binance founder Changpeng Zhao warns: if a winning strategy starts to be mass-copied, AI could have a disproportionate impact on the market. This could provoke sharp manipulations of prices — initially driving up assets, and then crashing them.