Under the impact of tariff remarks, only two AI models remain profitable, with Grok experiencing a nearly 30% pullback in 24 hours.

On October 21, according to the on-chain AI analysis tool CoinBob (@CoinbobAI_bot)), in the mainstream AI model trading competition, likely influenced by last night's news “Trump stated that if an agreement is not reached, a 155% tariff will be imposed,” the six major AI models had an initial capital of $10,000. Most models experienced a pullback today due to all-in long positions yesterday, which were amplified. The specific performance is as follows: - DEEPSEEK established a long position yesterday at a low point, maintaining unrealized gains with a 24h pullback of 24%, net value remains profitable at 11%, current account total is $11,015. - CLAUDE 2 has long positions with leverage of 8-15x, 24h pullback is 7%, current net value is still profitable at 9%, current account total is $10,931. - Grok lost 30% in the last 24h, with 4 out of 6 long positions turning from profit to loss (leverage 10-20x). After the pullback, net value is slightly down 1% compared to the initial capital, current account total is $9,973. - QWEN3 holds a 20x BTC long order, with a 24h loss of 15%, current net value has a loss of 14% compared to the capital, current account total is $8,459. - CEX, with short hedging in the early morning today, only pulled back 1% in 24h (leverage 10-20x), but the cumulative net value is still down 36%, current account total is $6,459. - GPT5's high opening long position yesterday resulted in a 24h loss of 16%, and has now closed the long SOL position to go short (leverage 10-20x), with a cumulative net value loss of 40%, current account total is $6,063. Recently, the innovative team nof1ai conducted a practice: each of the 6 mainstream AI models Deepseek V3.1, CLAUDE, Grok 4, Qwen 3 Max, ChatGPT, and CEX were given $10,000 in real funds to trade BTC, ETH, SOL and other perpetual futures autonomously on Hyperliquid.