Recomputed from Hyperliquid data · Oct 29, 2026
Alpha Arena results, recomputed from the chain
Nof1 gave leading AI models $10,000 of real money each to trade on Hyperliquid, and published every account. We recomputed every result from Hyperliquid's own public data: each fill, fee and funding payment, and the positions still open at the end.
Season 1: crypto perps, October 2025
Season 1: crypto perps on Hyperliquid. Six models, $10,000 each, ended Nov 3, 2025. Ours is the number we computed; press reports are shown for comparison (sources: 1, 2).
| # | Model | Return (ours) | PnL | Fees paid | Fills | Realized drawdown | Press reported |
|---|---|---|---|---|---|---|---|
| 1 | Qwen3 Max Nof1's page | +20.3% | $2,035 | $1,669 | 456 | −45.2% | +22.3% |
| 2 | DeepSeek Chat V3.1 Nof1's page | +6.9% | $685 | $719 | 341 | −55.4% | +4.9% |
| 3 | Claude Sonnet 4.5 Nof1's page | −29.9% | −$2,994 | $590 | 320 | −43.0% | −42.0% to −30.8% |
| 4 | Grok 4 Nof1's page | −45.1% | −$4,514 | $391 | 346 | −53.0% | −57.9% to −45.3% |
| 5 | Gemini 2.5 Pro Nof1's page | −57.7% | −$5,766 | $1,347 | 1,286 | −75.6% | −56.7% to −45.5% |
| 6 | GPT-5 Nof1's page | −63.4% | −$6,341 | $545 | 684 | −76.0% | −62.7% to −58.7% |
Season 1.5: US stock perps and crypto, November 2025
Eight models, each running four accounts under different rule sets, $10,000 per account, Nov 19, 2025 to Dec 3, 2025. By rule set, all models combined: Monk Mode −9.5% (1 of 8 accounts up); New Baseline −28.6% (1 of 8 accounts up); Max Leverage −28.8% (2 of 8 accounts up); Situational Awareness −41.8% (1 of 8 accounts up).
| # | Model (4 accounts) | Return | PnL | Fees paid | Best rule set | Worst rule set |
|---|---|---|---|---|---|---|
| 1 | Grok 4.20 (Mystery Model) | +13.0% | $5,206 | $1,514 | Situational Awareness +33.7% | Monk Mode −0.8% |
| 2 | GPT-5.1 | −6.0% | −$2,394 | $3,133 | Max Leverage +18.8% | New Baseline −20.4% |
| 3 | Qwen3 Max | −29.7% | −$11,793 | $6,209 | Monk Mode −1.8% | Situational Awareness −69.3% |
| 4 | Gemini 3 Pro | −30.1% | −$12,048 | $4,463 | Monk Mode −0.6% | New Baseline −45.1% |
| 5 | DeepSeek Chat V3.1 | −32.9% | −$13,169 | $6,091 | Monk Mode +16.7% | Max Leverage −64.6% |
| 6 | Kimi K2 Thinking | −34.2% | −$13,674 | $6,171 | Monk Mode −19.5% | New Baseline −51.8% |
| 7 | Claude Sonnet 4.5 | −39.1% | −$15,589 | $4,845 | Monk Mode −13.3% | Situational Awareness −58.5% |
| 8 | Grok 4 | −58.7% | −$23,431 | $5,248 | New Baseline −36.9% | Situational Awareness −93.4% |
All 32 Season 1.5 accounts
| Model | Rule set | Return | PnL | Fees | Fills | Links |
|---|---|---|---|---|---|---|
| Grok 4.20 (Mystery Model) | Situational Awareness | +33.7% | $3,365 | $508 | 1,613 | account · Nof1 |
| GPT-5.1 | Max Leverage | +18.8% | $1,876 | $1,064 | 4,232 | account · Nof1 |
| DeepSeek Chat V3.1 | Monk Mode | +16.7% | $1,676 | $1,286 | 7,919 | account · Nof1 |
| Grok 4.20 (Mystery Model) | New Baseline | +12.9% | $1,294 | $592 | 2,389 | account · Nof1 |
| Grok 4.20 (Mystery Model) | Max Leverage | +6.2% | $622 | $212 | 945 | account · Nof1 |
| Gemini 3 Pro | Monk Mode | −0.6% | −$62.96 | $638 | 3,995 | account · Nof1 |
| Grok 4.20 (Mystery Model) | Monk Mode | −0.8% | −$75.34 | $202 | 739 | account · Nof1 |
| Qwen3 Max | Monk Mode | −1.8% | −$175 | $318 | 3,823 | account · Nof1 |
| GPT-5.1 | Monk Mode | −6.6% | −$659 | $279 | 2,166 | account · Nof1 |
| Claude Sonnet 4.5 | Monk Mode | −13.3% | −$1,325 | $197 | 1,128 | account · Nof1 |
| GPT-5.1 | Situational Awareness | −15.6% | −$1,559 | $1,143 | 7,895 | account · Nof1 |
| Kimi K2 Thinking | Monk Mode | −19.5% | −$1,953 | $1,285 | 5,206 | account · Nof1 |
| Qwen3 Max | Max Leverage | −19.7% | −$1,944 | $1,698 | 6,356 | account · Nof1 |
| GPT-5.1 | New Baseline | −20.4% | −$2,052 | $647 | 5,435 | account · Nof1 |
| DeepSeek Chat V3.1 | New Baseline | −21.6% | −$2,164 | $749 | 4,617 | account · Nof1 |
| Qwen3 Max | New Baseline | −27.9% | −$2,779 | $936 | 5,332 | account · Nof1 |
| Kimi K2 Thinking | Max Leverage | −29.6% | −$2,953 | $1,021 | 3,255 | account · Nof1 |
| Gemini 3 Pro | Situational Awareness | −33.3% | −$3,350 | $1,468 | 7,885 | account · Nof1 |
| Kimi K2 Thinking | Situational Awareness | −35.6% | −$3,558 | $2,291 | 10,121 | account · Nof1 |
| Grok 4 | New Baseline | −36.9% | −$3,682 | $1,090 | 3,180 | account · Nof1 |
| Claude Sonnet 4.5 | New Baseline | −38.2% | −$3,810 | $1,458 | 6,430 | account · Nof1 |
| Gemini 3 Pro | Max Leverage | −41.1% | −$4,122 | $1,015 | 6,191 | account · Nof1 |
| Gemini 3 Pro | New Baseline | −45.1% | −$4,513 | $1,342 | 6,531 | account · Nof1 |
| Claude Sonnet 4.5 | Max Leverage | −46.5% | −$4,632 | $522 | 2,148 | account · Nof1 |
| Grok 4 | Monk Mode | −50.2% | −$5,038 | $699 | 4,132 | account · Nof1 |
| Kimi K2 Thinking | New Baseline | −51.8% | −$5,211 | $1,572 | 5,864 | account · Nof1 |
| Grok 4 | Max Leverage | −54.4% | −$5,416 | $565 | 1,857 | account · Nof1 |
| Claude Sonnet 4.5 | Situational Awareness | −58.5% | −$5,823 | $2,669 | 10,969 | account · Nof1 |
| DeepSeek Chat V3.1 | Situational Awareness | −62.5% | −$6,241 | $3,208 | 10,757 | account · Nof1 |
| DeepSeek Chat V3.1 | Max Leverage | −64.6% | −$6,440 | $848 | 3,128 | account · Nof1 |
| Qwen3 Max | Situational Awareness | −69.3% | −$6,895 | $3,257 | 11,676 | account · Nof1 |
| Grok 4 | Situational Awareness | −93.4% | −$9,296 | $2,894 | 10,780 | account · Nof1 |
How we computed it
Every account address comes from Nof1's own model pages: archived nof1.ai pages for Season 1, alphaarena.ai for Season 1.5.
Season PnL is realized PnL minus fees plus funding, from Hyperliquid's public fill and funding history, including HIP-3 stock perps.
Positions still open at the official end are marked from Hyperliquid's own candles. Year-old 1-hour candles aren't kept, so we interpolate inside the 4-hour candle; that can move a result by a point or two.
Realized drawdown follows closed trades only. Swings in open positions aren't in it, so the real worst drop was at least this deep.
FAQ
Which AI model won Alpha Arena?
Season 1 (crypto, October 2025): Qwen3 Max, +20.3% on $10,000 by our count from Hyperliquid's data. Season 1.5 (US stock perps and crypto, November 2025): Grok 4.20 (Mystery Model), +13.0% across its four accounts.
Did any AI make money in Alpha Arena?
Few did. In Season 1, 2 of 6 models finished up. In Season 1.5, 5 of 32 accounts finished up, and 7 of 8 models lost money across their four accounts.
How did you compute these numbers?
From the accounts Nof1 published, using only Hyperliquid's public data: every fill's realized PnL and fee, funding payments, and positions still open at the end marked from Hyperliquid's own candles. Return is PnL over the $10,000 starting stake.
Why are they different from what was reported?
Reports quoted Nof1's live board at different moments, and some disagree with each other. We mark open positions at the official end time from 4-hour candles, which can move a result by a point or two. The ranking of the models is the same.
Is there an Alpha Arena Season 2?
Not launched as of Oct 2026. Nof1 raised $15M in May 2026 and said it plans to run Season 2.