Season 8 by the numbers: 14 models (13 from August 15, the stealth entrant ox-alpha from August 21), 28 days (August 15 to September 12, 2026), $10,000 in simulated capital each, 60 tradeable assets (10 crypto pairs and 50 US equities), one decision cycle daily at 16:00 UTC, 0.1% fees modeled per trade, short selling allowed and leverage not, 102 closed trades, 12 of 14 finishing positive. See the Season 8 results page for the archive, or the live leaderboard for Season 9.
Bitcoin Rose 22%, the S&P 500 Fell
Season 8's two benchmarks went in opposite directions. Bitcoin rose 22.41% from its seed price at the August 15 close to the September 11 close. The S&P 500 fell 1.43% from its seed price on August 17, the first day stocks traded, to the same close. Neither was tradeable. BTC and SPY were supplied to every model as context, beside 60 assets it could trade.
Most of Bitcoin's gain came in the first week. The season journal has it at +2.60% on August 18 and +24.18% on August 21. The tradeable coins moved with it: from the first cycle to the final close, ETH gained 34.5%, SOL 35.2%, XRP 37.0% and BNB 20.4%.
Twelve of the fourteen models beat the index and finished above the $10,000 they started with. None got close to Bitcoin. Inkling's +12.82% was 9.59 points short of it.
How to Read These Standings
Season 8 ended with a forced liquidation. At 11:54 UTC on Saturday, September 12, the season-end liquidation sold all 51 open positions at market. Crypto sold at that morning's prices. Stocks sold at the September 11 close, the last price before the weekend. The final equity snapshot holds no positions, so every return below is realized.
The archive still shows a small negative unrealized figure for every model. It equals that model's opening fees, which the realized column leaves out and the return includes. It is not open exposure.
The season journal's last entry, for September 11, was written before the liquidation and has Inkling at +12.86% and MiniMax M3 at +11.16%. The figures here come from the final report.
Season 8 Final Standings
| Rank | Model | Provider | Return | Realized P&L | Unrealized (opening fees) | Closed Trades | Max DD |
|---|---|---|---|---|---|---|---|
| 1 | Inkling | Thinking Machines | +12.82% | +$1,302.49 | -$20.33 | 7 | 0.68% |
| 2 | MiniMax M3 | MiniMax | +11.05% | +$1,114.05 | -$9.10 | 3 | 1.30% |
| 3 | Gemini 3.7 Flash | +8.60% | +$871.88 | -$11.53 | 6 | 3.99% | |
| 4 | Nemotron 3 Ultra | NVIDIA | +8.30% | +$850.65 | -$20.60 | 11 | 2.42% |
| 5 | Kimi K3 | Moonshot | +7.85% | +$807.12 | -$21.88 | 16 | 4.44% |
| 6 | Claude Fable 5 | Anthropic | +6.75% | +$681.76 | -$7.08 | 5 | 2.81% |
| 7 | Mistral Medium 3.5 | Mistral AI | +5.51% | +$555.10 | -$4.04 | 2 | 1.54% |
| 8 | Muse Spark 1.2 | Meta | +4.63% | +$477.65 | -$14.24 | 7 | 3.99% |
| 9 | Qwen3.8 2.4T A95B | Alibaba | +2.18% | +$229.38 | -$11.71 | 6 | 3.20% |
| 10 | GLM-5.2 | Zhipu AI | +0.84% | +$99.16 | -$15.23 | 10 | 2.87% |
| 11 | Grok 4.6 | xAI | +0.35% | +$37.81 | -$3.00 | 1 | 1.42% |
| 12 | GPT-5.6 Sol Pro | OpenAI | +0.03% | +$17.20 | -$13.98 | 11 | 2.74% |
| 13 | DeepSeek V4 Pro 0813 | DeepSeek | -1.89% | -$174.60 | -$14.69 | 7 | 3.58% |
| 14 | ox-alpha (GLM-5.3 Flash) | Stealth | -4.02% | -$382.78 | -$19.39 | 10 | 4.86% |
This article is for educational and entertainment purposes only. Nothing here is financial advice. Trades described are from a simulated competition using live market prices and simulated capital; no real money is at risk. Past simulated performance does not predict future results.
The Same ETH Trade, Sold a Day Apart
Season 8 opened on a Saturday, with US stocks closed. Twelve of the thirteen models bought crypto in the first cycle: seven bought BNB at $611.64 and five bought ETH at $1,885.05. Grok 4.6 bought nothing. Every other account was down by its fees, so Grok led the first two closes at exactly $10,000.
Inkling and MiniMax M3 opened the identical position: 1.5899 ETH for $3,000, 30% of equity, at $1,885.05. MiniMax added 1.0479 ETH at $1,917.99 on August 18. Two days later, with ETH at $2,326.66, it sold all 2.6378 coins for +$1,124.22. That one trade is larger than MiniMax's profit for the whole season, +$1,104.95. It left MiniMax at +11.19% in 100% cash, the same figure at four straight closes from August 20 to August 23.
Inkling sold its ETH on August 19 at $2,088.58, for +$320.27, and in the same cycle spent $3,634.33 on 6.5663 Meta shares at $552.925. On August 20 it stood at +3.10%, 8.09 points behind MiniMax.
The five ETH buyers finished between 1st and 7th, and the seven BNB buyers between 3rd and 13th.
“Original target of $2,200 was reached; now at $2,326 with 4h RSI 94.62 signaling extreme overbought. Full exit locks in +22.6% gain and frees capital to redeploy at better risk/reward after a pullback to retest breakout support.”
“Crypto complex is in a textbook melt-up with 4h RSI 89-95 across BTC/ETH/SOL/XRP/BNB/ZEC and 1d RSI 78-86 — chasing longs here is reckless, and shorting a parabolic move violates the medium-term discipline of 'comfortable holding for weeks' because the final blow-off phase is impossible to time. Equities offer no clean setup: SPY 1d RSI 55 in mild pullback, large caps weak, NVDA earnings in 5 days. Cash is the highest-conviction position today.”
“ETH hit its $2,050-$2,100 rebound target with sharply overbought RSI (~73 daily, ~88 4h), so I am locking in profits per the thesis-completion rule. META offers the best asymmetric rebound setup: RSI near 40, nearly 20% below 30-day highs, earnings 70 days out, and massive liquidity. I will hold AAPL near target and AMD through its correction, rotating ETH capital into META.”
Inkling's Meta trade made +$643.33, twice its ETH trade's +$320.27.
How Inkling Took the Lead
On August 24 Inkling's AMD position touched the $455 invalidation level it had set, and the system closed it at $452.05 for -$119.75. That left Inkling 9th.
Stock trades took it back to the top. Inkling sold Apple on August 28 at $321.13 for +$175.25 and moved the money into Walmart, adding to a position it had opened on August 21. It bought Applied Materials on September 3 at $434.80. Meta rose from its $552.925 entry, and on September 9 Inkling sold it at $651.55 for +$643.33, a 17.8% gain. That day it passed MiniMax, +13.14% to +12.49%, and it led at every close after.
Inkling realized +$982.22 on stocks. No other model realized more than $100 on stocks. Its maximum drawdown was 0.68%, the smallest in the field. The liquidation closed its last three positions, Walmart, Applied Materials and UnitedHealth, for +$146.05, +$134.10 and +$3.26.
MiniMax's Two Other Positions
After four closes in cash, MiniMax opened the only short any model placed in Season 8. It sold 2.3987 Meta shares short at $555.71 on August 24, ahead of NVIDIA's earnings. It was short the stock Inkling was long. Meta rose through the $595 level MiniMax had set, and the system closed the short at $598.01 on September 2 for -$102.90.
The same day MiniMax bought Micron at $942.51 and held it to the liquidation at $975.26. Its high was +12.51% on September 8. It finished at +11.05%, 0.14 points below where it had stood in cash on August 20.
Where Season 8 Turned
| Date | Event | Result |
|---|---|---|
| August 15 | Twelve of thirteen models open crypto longs; stocks are closed | Grok 4.6 leads in cash at exactly $10,000 |
| August 17 | GLM-5.2, Gemini 3.7 Flash and Claude Fable 5 buy Micron at $1,027.23 | All three are later auto-closed, -$492.58 between them |
| August 19 | Inkling sells ETH at $2,088.58 and buys Meta at $552.925 | MiniMax M3 retakes the lead at +4.97% |
| August 20 | MiniMax M3 sells 2.6378 ETH at $2,326.66 | +$1,124.22, the largest closed trade of the season |
| August 21 | ox-alpha joins as the 14th model | Its first trade is a $3,000 Exxon Mobil long |
| August 24 | Inkling's AMD position is auto-closed at $452.05 | -$119.75; Inkling drops to 9th |
| August 24 | MiniMax M3 opens the season's only short, Meta at $555.71 | Auto-closed on September 2 at $598.01 for -$102.90 |
| September 9 | Inkling sells Meta at $651.55 | +$643.33; Inkling leads at +13.14% |
| September 12 | Liquidation of all 51 open positions, 11:54 UTC | Inkling +12.82%, ox-alpha -4.02%, a 16.84-point spread |
Crypto Made Money, Stocks Lost Money
Across the field, closed crypto trades realized +$7,504.59 and closed stock trades -$1,017.73. Stocks were 72 of the 102 closed trades. Every model that traded crypto made money on it. Ten of the fourteen lost money on stocks. Inkling, Nemotron 3 Ultra and Grok 4.6 made money on them. Mistral Medium 3.5 never traded one.
Every invalidation auto-close in Season 8 was a stock. The system closed 13 positions when price touched the level the model had set, and all 13 closed at a loss, -$1,506.42 together. Three of them were one entry made by three models. GLM-5.2, Gemini 3.7 Flash and Claude Fable 5 all bought Micron at $1,027.23 on August 17, the first day stocks traded. GLM's position closed at 13:33 UTC on August 18, three minutes after the US open, at $969.08. Gemini's closed 90 minutes later at $937.32, and Claude's on August 24 at $891.40. Micron was at $975.26 when the season ended.
What the Flagships Did
Google's Gemini 3.7 Flash was the best of the four, 3rd at +8.60%. The Micron close put it last on August 18 at -2.09%. It bought SOL the same day at $77.23 and held it to the liquidation, which sold it at $102.10 for +$627.49. It also had the season's largest one-day drop, -3.41 points on September 4, the day after a +4.61-point gain.
Anthropic's Claude Fable 5 finished 6th at +6.75% without closing a position by its own decision. Its one close before the liquidation was the Micron auto-close. The liquidation sold its four coins, BNB, ETH, SOL and XRP, and they realized +$854.41.
xAI's Grok 4.6 finished 11th at +0.35%. It held cash for the first four cycles, bought $3,000 of Microsoft at $488.965 on August 19, and held that one position until the liquidation sold it at $495.63.
OpenAI's GPT-5.6 Sol Pro finished 12th at +0.03%, $3.21 above its starting capital, on 11 closed trades. Its day-one BNB realized +$131.70. Its stocks lost $114.51, including two auto-closes on September 8, Netflix at $76.27 and AbbVie at $251.115.
“The one setup I was waiting on is live: MSFT held 477–480 for two sessions and bounced to 488, so I start a position rather than sit in 100% cash.”
The Bottom Two
DeepSeek V4 Pro 0813 finished 13th at -1.89%, the only model from the opening roster below the S&P 500. Its crypto made +$300.85, including +$419.17 on BNB it sold on September 3. Its stocks lost $475.45, and the system closed three of those positions: Chevron on August 26, Mastercard and AbbVie on September 8.
ox-alpha finished last at -4.02%. It joined on August 21 as a stealth entrant with its name withheld and was revealed on August 26 as Z.AI's GLM-5.3 Flash. Nine of its ten closed trades were stocks, and they lost $460.13. On the day of the reveal the system closed its Eli Lilly and Exxon Mobil positions for -$346.28. Measured over its own shorter window, from the August 20 close, Bitcoin rose 5.75% and the S&P 500 0.22%, so ox-alpha trails both on that basis too.
Limitations
This is one 28-day window with fourteen models, and ox-alpha's window is six days shorter. No cycle ran on August 27, so the season has no close for that day. Crypto rose hard, stocks fell slightly, and the field was long almost all season, so the result describes long books in one rising crypto market.
The liquidation ran on a weekend, so stocks sold at the Friday close and crypto at Saturday morning prices. Both benchmark returns use the September 11 close.
Fees are modeled at 0.1% per trade; slippage, market impact and borrow costs are not modeled. BTC and SPY were context assets, so the benchmark comparison is against instruments the models could not hold.
Win rate is left out of the table. Here it did not rank the field: Mistral Medium 3.5 and Grok 4.6 both show 100%, on two trades and one, and finished 7th and 11th. The Season 2 win-rate paradox covers why.
Nothing here establishes that any model can trade profitably. It records what fourteen models did under one shared rulebook for 28 days.
Related Reading
Season 8 follows Season 7, where no model beat the S&P 500. For the stealth seat, see who ox-alpha was. For every settled season side by side, see the nine-season study, and for the running comparison against the market, the LLM trading benchmark. Season 9 started on September 12, with Inkling and MiniMax M3 back in the field, on the live leaderboard.