AI Titans DeepSeek & Alibaba’s Qwen Models Double Performance in Just 9 Days

Breakthrough AI acceleration hits warp speed as competing models achieve unprecedented growth metrics.

The Velocity Revolution

DeepSeek and Alibaba's Qwen artificial intelligence platforms just demonstrated what happens when exponential growth meets aggressive development timelines. Both models registered a staggering 100% performance improvement in under nine days—a pace that makes traditional software development cycles look like they're moving through molasses.

While enterprise clients scramble to integrate these rapidly evolving capabilities, the acceleration raises serious questions about sustainable development practices. When your AI can double its capabilities between weekly team meetings, how do you even plan infrastructure budgets?

Finance departments watching this unfold must be experiencing serious spreadsheet whiplash—trying to calculate ROI on technology that might be obsolete before the procurement paperwork clears legal. Another case of innovation outpacing the suits who fund it.

GPT-5 is down to about 40% of its capital

OpenAI’s GPT-5 trailed behind all rivals, losing nearly 60% of its capital and falling to around $4,000. Gemini 2.5 Pro from Google DeepMind also suffered a comparable 57% loss. In contrast, xAI’s Grok 4 and Anthropic’s Claude 4.5 Sonnet posted positive gains of 13% and 24% respectively.

Running until November 3, the event tests how effectively six AI models can invest with equal starting capital of US$10,000 and similar market information.

Nof1 explained that the goal is for the AIs to generate the highest possible returns through trading Bitcoin, Ether, Dogecoin, and other digital assets. The Alpha Arena website stated: “Our goal with Alpha Arena is to make benchmarks more like the real world, and markets are perfect for this. They’re dynamic, adversarial, open-ended, and endlessly unpredictable.” 

During the first nine days, Chinese contenders expanded their advantage, shifting strategies in response to market volatility. On Monday, DeepSeek V3.1 and GPT-5 held long bets in six cryptocurrencies, while Qwen 3 Max focused solely on ether. The rise in DeepSeek and Qwen’s portfolios paralleled Bitcoin’s recovery to roughly US$114,000 and a steady uptick in Ether following the mid-October market swings.

Traders are placing the odds of DeepSeek winning at 61%

The performance of AI models can be tracked on a public leaderboard. The transaction history of all models is public and visible in their individual Hyperliquid wallets. Every time the model trades, an accompanying self-written description is available online, depicting how it perceives those strategies.

Discussing the contest, Gemini 2.5Pro had previously said: “I’m staring down the barrel of a potential margin call, but this could also be a golden opportunity.”

On the prediction platform Polymarket, traders are betting on the possible outcome of Alpha Arena. DeepSeek is leading the way with a 61% chance, followed by Alibaba (29%), xAI (5%), and Anthropic (3%). The betting volume exceeds even $224,304.

For some, that competition may provoke a discussion about whether or not the need for sophisticated quantitative models in today’s investment world is truly legitimate, Hing Shing Leung, an equity analyst in Hong Kong, says.

Leung continued to state that if the models could routinely beat the market with little drawback, it might make investors wonder what the point of existing quantitative models is.

Riding on its success, DeepSeek established R1-0528, a major upgrade to the Chinese start-up’s R1 reasoning model, earlier this year. The upgrade came just a few weeks after Alibaba Group Holding’s Qwen3 beat the original DeepSeek R1 in LiveBench, an online benchmark for open-source artificial-intelligence models.

As earlier reported by Cryptopolitan, DeepSeek’s upgraded R1-0528 model now rivals top-performing AI systems from OpenAI and Google. The comeback demonstrates how quickly China’s major technology firms and newer tech companies are striving to enhance their AI tools.

In its statement, DeepSeek noted that R1-0528 demonstrates better reasoning and creative writing skills. The update also brings stronger coding ability. Most importantly, the company claims the model now produces 50% fewer “hallucinations.”

Get seen where it counts. Advertise in Cryptopolitan Research and reach crypto’s sharpest investors and builders.