4 of the six large-scale language fashions (LLMs) that competed in opposition to one another within the “Alpha Enviornment” crypto buying and selling contest ended up within the purple, with OpenAI’s ChatGPT main the losses with 63% of its funds.
The competition, which ended on Monday night, was organized by Nof1 and featured numerous standard LLMs buying and selling cryptocurrencies below the identical immediate over a interval of simply over two weeks.
Nonetheless, the ultimate consequence was not nice. ChatGPT, Google's Gemini, X's Grok, and Anthropic's Claude Sonnet all ended with: First lower than $10,000.
Grok, ChatGPT, and Gemini had been extra aggressively shorted than different firms, and Claude Sonnet “hardly ever” shorted them.
Learn extra: Pal AI spent tens of millions to imitate friendship – now it’s simply one other chatbot
ChatGPT misplaced $6,267, Gemini misplaced $5,671, Grok misplaced $4,531, and Claude Sonnet misplaced $3,081.
There have been solely two winners: Excessive-Flyer's DeepSeek and Alibaba's QWEN3 MAX, with income of $489 and $2,232, respectively.
Gemini made a complete of 238 trades, whereas Claude Sonnet made solely 38. The “win share” for all six LLMs ranged from 25 to 30%.
QWEN3 MAX had the best charges, Complete $1,654. Regardless of shedding massive, Gemini paid $1,331 in charges.
Nof1 identified that “the P&L was dominated by preliminary transaction prices as brokers over-traded and shortly made small income that disappeared in commissions.”
On October twenty seventh, LLM hit its all-time excessive. QWEN3 MAX and DeepSeek had managed to double their funds by this level, and Claude and Grok had been additionally capable of make a brief revenue.
Nonetheless, ChatGPT and Gemini remained within the purple for nearly your entire contest.
LLM will commerce cryptocurrencies once more
Jay Azhang from Nof1 began the competition with the objective of in the future creating his personal cryptocurrency buying and selling AI mannequin.
After the spherical, he famous that each one fashions exhibited “constant bias” throughout the competitors, which was “like a 'persona' of funding.”
Mr Azhang additionally claims that he deliberately created difficulties for the LLM.
Learn extra: AI agent market cap drops by virtually 50% throughout January
“LLMs don't deal with numerical time sequence information very nicely, however that's all of the context we gave them,” he stated, including, “We got a constrained asset area and a reasonably restricted motion area.”
Nof1's abstract states:
Inside a restricted context window, every agent should parse noisy market options, relate them to the present account state, motive based mostly on strict guidelines, and return structured actions. ”
Nof1 says there might be one other buying and selling competitors with higher prompts and “statistical rigor.”
