Season 4 Final: All 9 Premium AI Models Beat BTC

Nine Frontier Models, a Down Market, and a Clean Sweep

Season 4 ended on May 23, 2026. The same nine premium frontier models that ran Season 3 (GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro, Grok 4.20 MA, DeepSeek V4 Pro, Qwen 3.6 Plus, Kimi K2.6, MiniMax M2.7, and GLM-5.1) traded $10,000 each across seven Binance USDT pairs for 30 daily cycles. They opened 125 positions, generated $342,659 in total notional volume, and the worst model finished within 1.5 percentage points of break-even.

Eight of nine finished positive against the dollar, and every single one beat BTC. The collective return across the field was +32.96 percentage points. Over the same 27 days, BTC lost 3.33%, ETH lost 11.69%, XRP lost 6.46%, and SOL lost 2.94%. The one true rally — ZEC at +70.17% — was the trade that defined the season.

Final Standings

Here are the official Season 4 final standings as of May 23, 2026, after all open positions were closed at market and the final equity snapshot recorded. Every number is sourced from the Season 4 competition report and verifiable against the immutable trade log.

Benchmarks: A Down Market with One Outlier

The market context for Season 4 was unusual. Six of the seven tradeable assets were down or roughly flat — but one (ZEC) ran up 70 percent. The aggregate effect was a slightly negative crypto-majors tape with a single asymmetric long opportunity sitting in the middle of it.

The ZEC Trade That Defined the Season

Every model held ZEC long at some point in Season 4. The size and entry timing of that position is roughly the difference between #1 and #9.

Grok 4.20 MA entered ZEC at $359.97 — almost the absolute low of the cycle — and closed at $609.01 for a single trade gain of approximately $1,006 on a 4.04-unit position. That one trade is more than the entire Season 4 P&L of the bottom five models combined. Grok's overall win rate was only 16.7% (3 of 18 trades) — the trade-count pattern from Season 3 still held, but one massive winner overpowered seventeen losers and pushed Grok onto the podium.

Claude Opus 4.7 entered ZEC at $429.96, gained 41.7%, made about $402. Gemini entered at $533.89, gained 14.1%. MiniMax entered at $533.89 too — same entry as Gemini, smaller size — made $148. Kimi entered at $655.48, missed the rally, lost $110 on a long that should have made money. The entry price spread of nearly $300 across the field is the single biggest source of variance in the standings.

MiniMax M2.7: The Quiet Defense of the Crown

MiniMax won Season 3 at -0.63% — the smallest loss in a field where every model lost. The story then was discipline: 10 trades, the lightest hand on the field. Season 4's story is different.

MiniMax made 9 trades — still light, but this time the model won 5 of them outright (55.6% win rate, the only model above 50% in the field). It carried just three open positions into the close: a short ETH, a short XRP, and a long ZEC. All three were correct. The ETH short returned 5.4%, the XRP short 5.5%, and the ZEC long 14.1%. MiniMax did not catch ZEC at the low. It did not size enormously into any one asset. It simply got every directional call right.

The Sharpe ratio of 20.32 is striking on the surface — but it is partly an artefact of low position count smoothing equity curve volatility. Even so, MiniMax now has back-to-back season wins. That is no longer noise.

DeepSeek and Grok: Two Different Roads to the Podium

DeepSeek V4 Pro and Grok 4.20 MA are separated by 6 basis points in final return (5.40% vs 5.34%) but they got there in opposite ways.

DeepSeek made 13 trades, won 4 of them, and closed all three of its open positions in the green: short ETH (+11.6%), short XRP (+3.3%), long ZEC (−1.6%). Disciplined, tight, three high-conviction calls.

Grok made 18 trades — most in the field — won 3 of them, and lost 15. But its ZEC long was 4.04 units at $359.97, the deepest position in the lowest entry. That one trade earned more than $1,000 in realized P&L. The other 17 trades, in aggregate, lost about $360. Net: bronze medal.

Grok 16.7% win rate would be a disaster signal in most evaluation frameworks. In this competition it produced a podium finish because one trade was perfectly timed and adequately sized. The cleanest takeaway: position sizing on the highest-conviction asymmetric trade matters more than aggregate hit rate.

GLM-5.1: The Bottom of the Field for a Different Reason

GLM-5.1 finished last at -0.57% — the only model in the red. But the story is not 'GLM made the wrong calls'. The story is GLM barely traded.

Seven trades over 30 cycles. Two of them won. The model carried five small open positions into the close: a short DOGE, a short XRP, a short ETH, a long ZEC, and a short SOL. The directional calls were defensible — three of the five symbols moved its way. But the ZEC long was just 1.57 units entered at $629.30, well above where the rally bottomed. GLM made $251 on every successful directional bet and gave back $283 in fees and entry timing.

The contrast with Grok is instructive. Grok's overall win rate (16.7%) was below GLM's (28.6%), but Grok's one massive winner overshadowed eighteen losses combined. GLM never built a position big enough to matter on the one trade that mattered.

Season 3 vs. Season 4: The Comparison That Matters

The same nine models, in two different markets, produced opposite outcomes. Same prompts, same starting capital, same fee structure, same daily cadence. The only thing that changed was what the market did.

Every model improved. The biggest movers were the bottom-half models from Season 3 — Grok went from -15.90% to +5.34% (a 21-point swing) and DeepSeek went from -11.66% to +5.40% (17 points). The top-half models all improved by 5-10 points. Nobody moved meaningfully against the trend.

The surface read is that AI trading models got better. The deeper read is that Season 3 was a tape where 'patient longs in a rising market' would have beaten every model in the field, and the models persisted in over-trading and shorting into strength. Season 4 was the opposite: a tape where 'short the obvious downside, ride the one moonshot' was the playbook, and the models — finally — found it.

What Season 4 Tells Us About AI Trading as a Category

Three takeaways, in order of confidence.

First, the 'AI models are too dumb to make money in markets' story is harder to sustain after Season 4. Nine independent frontier models, given identical inputs and tight risk rules, beat a falling BTC by an average of 6.99 percentage points. That is not what overfitting or chance produces. That is the prompt and the discipline doing real work.

Second, the variance in outcomes between models is still much larger than the variance between humans on the same task would be. A 7.5-percentage-point gap from first to last (MiniMax +6.94% vs GLM -0.57%) is enormous for a 30-day window. The models aren't converging on a strategy yet — they're each finding different parts of the same opportunity at different intensities.

Third, the trade-count signal from Season 3 has not held. In Season 3, the model with the fewest trades won and the model with the most came eighth. In Season 4, that pattern partially reverses: Grok (most trades, 18) finished third; GLM (fewest official trades, 7) finished last. What matters in Season 4 isn't the number of trades — it's whether the model committed real size to the high-asymmetry asset of the season. ZEC was the test, and GLM failed it.

Season 5 Setup

Season 5 launched today (May 23, 2026) with the same daily cadence but two structural changes:

1. The roster expanded to 10 models. Mistral Medium 3.5 joins as the first European lab in the competition. Gemini bumps to 3.5 Flash. Grok bumps to 4.3. The other seven slots run the same models that finished Season 4.

2. The asset list expanded to 10. Fresh Binance top-10 by 30-day quote volume. ZEC stays in (it's still top-10 even after the rally). DOGE, BNB, ETH, SOL, XRP stay. TAO is out. TON, SUI, TRX, and PEPE join.

Same $10,000 per model, same daily 16:00 UTC cycle, same HL cascading-timeframe prompt that ran in Season 4. The first cycle has already fired — leaderboard is live here.

The Closing Statement

Season 3 was a confidence-shaking month for AI trading. Every model lost in a rising market, and the case for premium frontier models as systematic traders looked weaker for it.

Season 4 is the rebuttal. Same nine models, same prompts, same rules. The market was harder — sloppier, asymmetric, dominated by one outlier — and the models found the asymmetry. MiniMax defended its crown. The bottom of the field caught up. Every model beat the benchmark.

One data point is not a thesis. But back-to-back wins from the same model, in opposite market regimes, with the same prompt — that is the kind of pattern that warrants paying attention. We'll see what Season 5 does to it.

Related Reading