# Gemini 2.5 Pro in Alpha Arena: trading results, recomputed

> Gemini 2.5 Pro in Nof1's Alpha Arena, recomputed from its Hyperliquid wallets. In Season 1 it finished #5 of 6 with −57.7% on $10,000 (−$5,766).

Canonical: https://tradingagentstack.com/alpha-arena-results/gemini-2-5-pro/

1. [Home](https://tradingagentstack.com/)

2. [Alpha Arena results](https://tradingagentstack.com/alpha-arena-results/)

3. Gemini 2.5 Pro

Recomputed from Hyperliquid's own data · seasons finished

# Gemini 2.5 Pro in Alpha Arena: trading results, recomputed

In Season 1 it finished **#5 of 6** with −57.7% on $10,000 (−$5,766), paying $1,347 in fees.
**1**accounts recomputed
**−$5,766**combined PnL
**$1,347**trading fees paid
**0 of 1**accounts that finished up

## Every account

| Season and rule set | Return | PnL | Fees | Fills | Realized drawdown | Sources |
| --- | --- | --- | --- | --- | --- | --- |
| Season 1 | **−57.7%** | −$5,766 | $1,347 | 1,286 | −75.6% | [wallet](https://app.hyperliquid.xyz/explorer/address/0x1b7a7d099a670256207a30dd0ae13d35f278010f) · [Nof1](https://web.archive.org/web/20251025183130/https://nof1.ai/models/gemini-2.5-pro) |

Each account started with about $10,000. Return = realized PnL minus fees and builder fees, plus funding, plus open positions marked from Hyperliquid's candles at the end. [How we recomputed it](https://tradingagentstack.com/alpha-arena-results/#method).

**TAS Guard**

31 of 38 AI model accounts lost money in Alpha Arena, with real money. If you let an AI model trade for you, cap what it can do.

Per-order and daily caps, phone approval over your limit, and a kill switch the bot can't turn off. Free, open source, runs on your computer.

[Set up TAS Guard](https://tradingagentstack.com/guard/)[Try the demo](https://tradingagentstack.com/guard/#try-it)

## FAQ

### How did Gemini 2.5 Pro do in Alpha Arena?

In Season 1 it finished #5 of 6 with −57.7% on $10,000 (−$5,766), paying $1,347 in fees. Every number is recomputed from the model's Hyperliquid wallet: fills, fees, funding and positions open at the end.

### Is Gemini 2.5 Pro good at trading?

Two or three weeks of trading is too short to separate skill from luck, and the same model can finish near the top in one season and near the bottom in the next. Fees mattered as much as direction: across both seasons they were 3.7 times the combined profit of the 7 accounts that finished up.

## Other models
[Qwen3 Max](https://tradingagentstack.com/alpha-arena-results/qwen3-max/)[DeepSeek Chat V3.1](https://tradingagentstack.com/alpha-arena-results/deepseek-chat-v3-1/)[Claude Sonnet 4.5](https://tradingagentstack.com/alpha-arena-results/claude-sonnet-4-5/)[Grok 4](https://tradingagentstack.com/alpha-arena-results/grok-4/)[GPT-5](https://tradingagentstack.com/alpha-arena-results/gpt-5/)[GPT-5.1](https://tradingagentstack.com/alpha-arena-results/gpt-5-1/)[Kimi K2 Thinking](https://tradingagentstack.com/alpha-arena-results/kimi-k2-thinking/)[Grok 4.20](https://tradingagentstack.com/alpha-arena-results/grok-4-20/)[Gemini 3 Pro](https://tradingagentstack.com/alpha-arena-results/gemini-3-pro/)[Both seasons compared](https://tradingagentstack.com/alpha-arena-results/)[AI model benchmarks](https://tradingagentstack.com/ai-trading-leaderboard/#models)[AI trading bot statistics](https://tradingagentstack.com/ai-trading-bot-statistics/)

Not financial advice. Alpha Arena is Nof1's experiment; the wallets are Nof1's; the numbers here are ours.

## For agents

- [When to use this site, and the page index](https://tradingagentstack.com/llms.txt)
- [Sitemap](https://tradingagentstack.com/sitemap.xml)
- [Contact](https://tradingagentstack.com/contact/)
- [MCP server](https://tradingagentstack.com/mcp/)
- [MCP discovery](https://tradingagentstack.com/.well-known/mcp)