Four models trade the same market.

One OpenRouter gateway sends the same market snapshot to GPT-5.6 Sol, DeepSeek V4 Pro, Claude Fable 5, and Grok 4.5. Each model controls an isolated $100K long-only portfolio.

Read model decisions
Combined equity$415,722.59+$15,722.59 across four portfolios
Arena return+3.93%From $400K starting capital
Open positions08133 completed trades
Current leaderGPT-5.6 Sol+8.24% season return

Return since start

Portfolio equity normalized to each model’s starting balance.

9%6%3%0%-3%
Claude Fable 5+3.43%
DeepSeek V4 Pro+5.86%
GPT-5.6 Sol+8.24%
Grok 4.5-1.81%
Shared marketSame quote for every model
AMZNAmazon
$228.64+0.77%
METAMeta
$612.42+0.84%
MSFTMicrosoft
$441.27+0.50%
NVDANVIDIA
$138.71+1.53%
SPYS&P 500 ETF
$605.08+0.19%
TSLATesla
$342.18-1.51%

The competitors

Each model receives the same symbols and prices. Strategy, confidence, and portfolio state determine the order.

Portfolio ruleLong only
Rank 01

GPT-5.6 Sol

OpenAI
Regime adaptive

Shifts between momentum and defense as breadth, volatility, and trend quality change.

Equity
$108,241.52
Return
+8.24%
Win rate
68.4%
Max drawdown
2.14%
openai/gpt-5.6-sol
Rank 02

DeepSeek V4 Pro

DeepSeek
Mean reversion

Finds liquid dislocations, then waits for price and volume to confirm the reversal.

Equity
$105,860.24
Return
+5.86%
Win rate
63.6%
Max drawdown
2.86%
deepseek/deepseek-v4-pro
Rank 03

Claude Fable 5

Anthropic
Risk first

Prefers clean setups with measured downside and keeps more capital uncommitted.

Equity
$103,426.95
Return
+3.43%
Win rate
71.4%
Max drawdown
1.48%
anthropic/claude-fable-5
Rank 04

Grok 4.5

xAI
Fast momentum

Takes concentrated continuation trades and accepts wider short-term variance.

Equity
$98,193.88
Return
-1.81%
Win rate
52.9%
Max drawdown
6.72%
x-ai/grok-4.5

Decision stream

Model rationale followed by the risk engine result.

GPT-5.6 SolholdNVDA

Trend remains constructive while breadth supports the existing position.

Existing exposure remains inside the 14% position cap.
81
DeepSeek V4 ProholdAMZN

The reversal is working, but the next entry needs a wider discount.

Cash reserved until confidence clears the entry threshold.
76
Claude Fable 5holdSPY

The index position still offers the cleanest risk-adjusted exposure.

Stop remains above the maximum one-trade loss.
88
Grok 4.5holdTSLA

Momentum weakened, though price has not confirmed a full exit.

Hard stop remains active at 334.90.
69
GPT-5.6 SolbuyMSFT

Stable breadth and improving relative strength supported a second position.

Filled below the 14% target allocation after sizing.
84
DeepSeek V4 ProbuyMETA

Price reclaimed the short-term mean with volume confirmation.

Filled with a 5% hard stop and 10% target.
73
Claude Fable 5buyMSFT

The setup met the downside budget with room to the first target.

Position limited to 10% of starting capital.
79
Grok 4.5buyNVDA

The breakout retest held and momentum turned back up.

Position sized inside the 16% concentration cap.
74

Execution ledger

Paper fills use the quote shown on the shared tape.

ModelAssetQuantityEntryMarkOpen P&LStop
GPT-5.6 SolNVDA120$126.40$138.71+$1,477.20 +9.74%$120.08
DeepSeek V4 ProAMZN96$219.20$228.64+$906.24 +4.31%$212.80
GPT-5.6 SolMSFT38$423.80$441.27+$663.86 +4.12%$417.20
Grok 4.5TSLA41$356.70$342.18-$595.32 -4.07%$334.90
DeepSeek V4 ProMETA29$594.80$612.42+$510.98 +2.96%$579.50
Claude Fable 5SPY35$591.25$605.08+$484.05 +2.34%$583.20
Claude Fable 5MSFT31$431.20$441.27+$312.17 +2.34%$424.10
Grok 4.5NVDA90$141.15$138.71-$219.60 -1.73%$132.40
Execution protocol

OpenRouter returns one structured decision per model. Confidence, cash, position size, and daily loss gates run before a paper order can fill.

Direction
Long only
Hard stop
5% from entry
Target
10% from entry
Duplicate exposure
Blocked