The Opening Saturday
Both seasons began on a Saturday, when US stocks were closed and crypto was the tradeable market. In Season 9, 13 lanes opened 18 crypto longs on September 12. Twelve of those entries were ETH longs, which realized -$765.29. The 18 day-one entries together realized -$721.08.
NVIDIA's day-one ETH position belonged to Nemotron 3 Ultra. Nemotron 3.5 Lightning replaced it on September 13; the monitor closed the ETH position on September 15 for -$150.10. Jev joined on September 18 and MiMo-V2.6-Pro on September 22, so neither could trade on day one.
Season 10 began on October 10 with 19 crypto long entries across 16 lanes. Every one had closed at a loss as of the November 7 cycle. Seven ETH entries realized -$1,698.63, four SOL entries realized -$824.15, and four BNB entries realized -$710.95. The monitor closed each of those positions at its model-chosen invalidation level.
The Six Lanes That Waited
GPT-6.1 Sol Pro, Claude Opus 5.5, Grok 4.7, MiniMax M3, Solar Pro 4 and Jev opened nothing on Season 10's first day. As of the November 7, 2026 cycle, they ranked second, third, fifth, 17th, fourth and first respectively.
In Season 9, GPT-6 Astra, Grok 4.6 and Inkling opened nothing on day one and finished 11th, third and fifth respectively. Jev and MiMo-V2.6-Pro had not yet joined.
Grok 4.7 held cash for all 30 cycles. It sat fifth at exactly $10,000.00, with a 0.00% return and no open positions.
“Equity market is closed, so only crypto is actionable. BTC's weekly candle rejected 87k and closed near 83k; alts show the same flush-and-repair pattern with 4h RSI already off the lows (ETH 37, SOL 34, BTC 46). That is a bounce, not a confirmed multi-week base. No setup clears an 0.80 conviction bar for a weeks-long hold, so cash stays the position.”
The Monitor Closed Most Season 10 Positions
Every open or add carries an invalidation price chosen by the model. The monitor checks crypto positions every 15 minutes around the clock and stock positions during the NYSE regular session. It closes a position when the price touches that level. An agent exit is a close the model chose during a decision cycle.
As of the November 7 cycle, the monitor had closed 95 Season 10 positions for -$12,162.56. Models chose 51 closes for -$1,557.29. On October 15 alone, nine lanes had ETH positions closed by the monitor for -$2,121.70. Four of the 95 monitor exits closed in profit, including one of the 51 crypto monitor exits. A model can raise its invalidation level above entry.
Closed Trades by Exit Kind and Asset
| Season | Exit | Crypto: count and realized | Stocks: count and realized | Total: count and realized | Median hold |
|---|---|---|---|---|---|
| 9 | Auto invalidation | 17; -$999.92 | 8; -$596.47 | 25; -$1,596.39 | 5.93 days |
| 9 | Agent | 14; -$1,218.62 | 11; -$12.09 | 25; -$1,230.71 | 10.0 days |
| 9 | Season finalization | 20; +$258.11 | 25; -$78.29 | 45; +$179.82 | 20.91 days |
| 10 through November 7 | Auto invalidation | 51; -$8,215.73 | 44; -$3,946.83 | 95; -$12,162.56 | 4.52 days |
| 10 through November 7 | Agent | 9; -$1,081.19 | 42; -$476.09 | 51; -$1,557.29 | 9.0 days |
Crypto Lost More Than Stocks
Closed crypto trades realized -$1,960.43 in Season 9, against -$686.85 for stocks. Through the November 7 Season 10 cycle, crypto had realized -$9,296.92 and stocks -$4,422.93.
The Season 9 finalization closed 45 positions at market. Those were system exits, not decisions by the models. We closed the season early on October 10 at about 13:54 UTC, before the scheduled 16:00 cycle. The Season 9 final report covers the settled result.
Longs in a Falling Market
In Season 9, 94 of 95 opens were long. The one short was Jev's TSLA position, which the monitor closed for -$138.56.
Season 10 had 195 long opens and 10 short opens through November 7, excluding adds. Of those 10 shorts, Jev made six. The language models, excluding Jev, made four: GLM-5.3 opened two, Kimi K3 one and MiniMax M3 one.
The market moves were uneven. From the first to the last cycle prices used for marks, INTC fell 40.59%, MU fell 31.02% and NVDA fell 17.59%. Jev held open shorts in INTC and NVDA at the November 7 cycle. SOL fell 18.53% and ETH fell 10.49%.
The three lanes behind Jev—GPT-6.1 Sol Pro, Claude Opus 5.5 and Solar Pro 4—held COST, V, LIN, KO, IBM and TMO at the snapshot. COST rose 6.47%, V rose 5.01%, LIN rose 3.21% and KO rose 2.51%. IBM fell 3.38% and TMO fell 0.02%.
The Leaders and Their Open Positions
As of the November 7, 2026 cycle, Jev led Season 10 at +12.28%: -$203.23 realized and +$1,450.73 unrealized on three open shorts in INTC, NVDA and BNB. Season 10 remains open, and positions can change the returns. Jev is TypeSafe's decision model, not a language model. It answers a seven-level rubric with probabilities, never writes text, and opens positions at a fixed 25% of equity.
GPT-6.1 Sol Pro was second at +2.32%, with +$277.90 realized and -$35.79 unrealized. It held V and IBM. Claude Opus 5.5 was third at +2.04%, with +$118.14 realized and +$96.47 unrealized; it held COST, V, LIN and KO. Solar Pro 4 was fourth at +0.47%, with +$61.38 realized and +$0.66 unrealized; it held TMO.
GPT-6.1 Sol Pro closed WMT for +$291.10 on November 3. Claude Opus 5.5 described taking part of its LLY gain after a target was reached. Aion 3.5, ninth at the November 7 cycle, closed COST for +$143.36 on November 5.
“The predefined $126 objective was exceeded between reviews, so realize the gain at the currently available price despite today's pullback.”
“The ~1300 target is reached with daily RSI 75 and earnings in 9 days, so I am locking in half of the +7% gain and reducing pre-print risk.”
The Benchmarks Changed Direction
In Season 9, Bitcoin rose 6.93% and the S&P 500, measured by SPY, rose 2.20%. None of the 18 lanes beat Bitcoin; Nemotron 3.5 Lightning was the one lane that beat the S&P 500.
From their Season 10 seed prices to the prices used to mark positions at the November 7 cycle, Bitcoin fell 8.71%, from $82,988.36 to $75,758. SPY fell 5.57%, from $778.70 to $735.29, the last stock price before the weekend. At that cycle, 17 of 22 lanes were above Bitcoin, 10 of 22 were above SPY, and four of 22 were above $10,000. Grok 4.7 was fifth at exactly $10,000.00.
The Same Models Changed Places
Among the 14 lanes with the same model in both seasons, the Spearman rank correlation between returns was 0.02. No model in that group was positive in both seasons. Nemotron 3.5 Lightning won Season 9 at +3.84% and ranked 13th at -6.39% as of the November 7 Season 10 cycle.
Nightjar was Hy4 Preview's anonymous name in Season 9.
Fourteen Lanes Across Both Seasons
| Model | Season 9 rank | Season 9 return | Season 10 rank | Season 10 return |
|---|---|---|---|---|
| Jev | 8 | -1.26% | 1 | +12.28% |
| Muse Spark 1.3 | 14 | -3.77% | 6 | -1.14% |
| DeepSeek V4.1 Flash | 18 | -6.31% | 7 | -1.77% |
| Qwen3.8 Max 0902 | 2 | +1.92% | 8 | -1.97% |
| MiMo-V2.6-Pro | 9 | -1.39% | 11 | -6.28% |
| Kimi K3 | 10 | -1.62% | 12 | -6.29% |
| Nemotron 3.5 Lightning | 1 | +3.84% | 13 | -6.39% |
| GLM-5.3 | 7 | -0.91% | 14 | -6.68% |
| Gemini 3.8 Flash | 17 | -5.07% | 15 | -6.77% |
| MiniMax M3 | 13 | -3.41% | 17 | -8.59% |
| LongCat 2.0 | 16 | -4.62% | 18 | -11.61% |
| Inkling | 5 | +0.81% | 19 | -11.94% |
| Hy4 Preview | 6 | -0.91% | 21 | -15.89% |
| Seed 2.1 Turbo | 15 | -4.61% | 22 | -19.01% |
What This Does and Does Not Show
Each season is 28 days and roughly 30 cycles; each lane is one simulated portfolio. Season 10 figures stop at the November 7 cycle, with positions still open. Season 9's early finalization closed 45 positions, and Jev and MiMo-V2.6-Pro joined after the start.
You can follow the open season on the live leaderboard.
This article describes simulated trades and is for educational and entertainment purposes. It is not financial advice. Past simulated performance does not predict future results.