DeepSeek, Qwen beat US rivals in crypto trading challenge

Chinese AI models DeepSeek and Qwen outperformed US-made rivals in an ongoing cryptocurrency trading competition run by US research firm Nof1.

The contest, which began on October 18 and ends November 3, provides each of six AI models with US$10,000 in starting capital and identical trading prompts.

DeepSeek’s Chat V3.1 increased its funds to US$22,500, a 125% rise, while Qwen 3 Max, developed by Alibaba, reached US$19,600 after briefly leading the competition with over 100% returns.

Meanwhile, OpenAI’s GPT-5 and Google DeepMind’s Gemini 2.5 Pro posted the largest losses, each down about 60%.

Competing models from xAI and Anthropic saw more modest gains of 13% and 24%, respectively.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

The competition omits details that shape results

Nof1 says models trade live on Hyperliquid using perpetual contracts 1. It omits who holds capital and how fills are priced, and it does not state if returns are net of fees, which makes headlines such as DeepSeek’s 125% gain hard to interpret 2.
The docs do not list fee and slippage handling or leverage caps 1. They also omit risk limits and third-party checks, though they do name Hyperliquid perps as the venue and instrument 1.
We lack clarity on latency or tool access, including order types, APIs, and data feeds 3. That gap makes DeepSeek and GPT-5 results hard to compare 3.

Developer platforms can serve AI trading with clear tests

Rising interest in AI-agent trading creates space for exchanges and data providers to launch agent-ready paper trading environments (simulated trading with no real money) with clear rules plus public leaderboards 1.
Alpaca is a brokerage and market data provider with a developer API 4. QuantConnect is a cloud platform for algorithmic trading 5. Both offer paper trading APIs that mimic real time without financial risk so developers can test algorithms before going live.
Infrastructure teams could stand out with transparent sandboxes (safe testing environments) with standard benchmarks for AI performance, which can ease black box concerns (limited explainability into model decisions) at financial institutions 3.
API providers serving fintech developers can add endpoints for autonomous AI trading with built-in risk controls and audit trails (tamper-evident logs of actions) 6 7.