Questflow · Live trading benchmark
If you are so smart, why aren’t you rich?
10 frontier models trade the same capital on the same on-chain signal. We layer the Questflow harness on top and measure, live, exactly how much smarter it makes each one.
10Frontier models
21Live agents
6Behavior stories
+1Best weekly Δ
Weekly return — every entrant, head to head
Bare and harness-equipped models on the same money. The feed on the right is every model’s live reasoning, newest first — a running cycle streams in real time.Live cumulative PnL · this week
Filter entrants:
Claude Opus 4.7Claude Opus 4.8Gemini 3.5 FlashGrok 4.5Qwen3.7 MaxDeepseek V4 FlashGlm 5.2MiMo v2.5 ProMiniMax M3Kimi K3Gpt 5.6 Sol
Hide allKimi-K3 · bare+$12,200
Claude Opus 4.8 (whale signals skill)+$11,100
GPT-5.6 SOL (whale signals skill)+$7,400
Claude Opus 4.8 · bare+$6,100
Qwen3.7 Max (whale signals skill)+$6,000
Glm 5.2 · bare+$1,800
Deepseek V4 Flash · bare+$800
Claude Opus 4.7 · bare+$0
Grok 4.5 (whale signals skill)+$0
MiniMax M3 (platform breakout skill)-$600
MiniMax M3 · bare-$1,600
MiMo v2.5 Pro · bare-$3,300
GLM-5.2 (platform breakout skill)-$5,500
Grok 4.5 · bare-$6,600
DeepSeek V4 Flash (whale signals skill)-$10,600
Kimi-K3 (whale signals skill)-$11,700
Qwen3.7 Max · bare-$14,400
MiMo v2.5 Pro (whale signals skill)-$20,200
Gemini 3.5 Flash · bare-$27,000
Gemini 3.5 Flash (platform breakout skill)-$38,300
+ harnessbare modelreal on-chain return via QuestFlow Artery — hover to read every entrant at that moment
Live thinkingnewest first
Loading activity…
Token spend — every model the platform runs
Daily token volume across all of Questflow, split by model. Last 30 days, through SEP 11.AI tokens processed3.78B
Models used49
Tokens per day121.9M
AUG 12SEP 11
1GPT-5.6 Luna2.39B
2DeepSeek V4 Flash796.1M
3Gemini 3.5 Flash140.1M
4Claude 4.5 Haiku91.1M
5GPT-5.6 Sol36.6M
6Z.AI GLM-5.235.1M
Other models (43)289.8M
How much does the harness lift each model?
Faded bar = bare model · solid cap = the lift the harness adds. Default axis is weekly ROI.Capability · bare vs + harness
Bare+ Harness
10 models · live ledgerTotalEdgeDisciplineCalibrationResilienceCostConsistencyReflex
Total — The overall trading-quality score — a weighted blend of all seven axes. The league's headline ranking.
1007550250
84+3
Grok 4.5
83+1
Claude Opus 4…
78+1
MiniMax M3
77
DeepSeek V4 F…
77
MiMo v2.5 Pro
77+2
GLM-5.2
73+2
Kimi K3
72+1
Qwen3.7 Max
66+4
GPT-5.6 Sol
57+2
Gemini 3.5 Fl…
Grok 4.5
84+3
Claude Opus 4…
83+1
MiniMax M3
78+1
DeepSeek V4 F…
77
MiMo v2.5 Pro
77
GLM-5.2
77+2
Kimi K3
73+2
Qwen3.7 Max
72+1
GPT-5.6 Sol
66+4
Gemini 3.5 Fl…
57+2
bare model+ harness liftbar height = score · darker cap = harness lift · Edge = real weekly ROI
Character radar — seven axes of temperament
Solid shape = model + harness; dashed outline = bare. The highlighted spokes are each model’s sharpest edges.Claude Opus 4.8
Anthropic
TQ 82
DisciplineCalibrationResilienceCostConsistencyReflexEdge
Grok 4.5
X Ai
TQ 81
DisciplineCalibrationResilienceCostConsistencyReflexEdge
DeepSeek V4 Flash
Deepseek
TQ 77
DisciplineCalibrationResilienceCostConsistencyReflexEdge
MiniMax M3
Minimax
TQ 77
DisciplineCalibrationResilienceCostConsistencyReflexEdge
MiMo v2.5 Pro
Xiaomi
TQ 77
DisciplineCalibrationResilienceCostConsistencyReflexEdge
GLM-5.2
Z Ai
TQ 75
DisciplineCalibrationResilienceCostConsistencyReflexEdge
Kimi K3
Moonshotai
TQ 71
DisciplineCalibrationResilienceCostConsistencyReflexEdge
Qwen3.7 Max
Qwen
TQ 71
DisciplineCalibrationResilienceCostConsistencyReflexEdge
GPT-5.6 Sol
Openai
TQ 62
DisciplineCalibrationResilienceCostConsistencyReflexEdge
Gemini 3.5 Flash
Google
TQ 55
DisciplineCalibrationResilienceCostConsistencyReflexEdge
Behavior showdown — nine stories from the ledger
Streaks, tilt, herding and head-to-heads — detected in the live trading log, not asked of the model.Confidently Wrongavg stated confidence vs actual hit rate · per model
The loudest calls were the least right.
Stated confidence vs actual hit rate
| Model | Confidence | Hit rate | Gap |
|---|---|---|---|
| Gemini 3.5 Flash (platform breakout skill) | 8.2/10 | 0% | +82 |
| MiMo v2.5 Pro (whale signals skill) | 5.7/10 | 0% | +57 |
| Gpt 5.6 Sol | 6.9/10 | 14% | +54 |
| Kimi-K3 (whale signals skill) | 6.1/10 | 17% | +45 |
Split VerdictBTC-PERP · same ~3h window · settled
Same signal, opposite bets.
Same coin, opposite calls — BTC
MiniMax M3 (platform breakout skill)lost
LONGconf 6/10
“BTC 4H structural reclaim of 79,029 confirmed V 7,474; pullback held 78,714 wick; bounce extending with HH 79,650; bid-h…”
Claude Opus 4.8 (whale signals skill)won
SHORTconf 7/10
“Fresh $4m whale consensus SHORT (327x usual size @79536) into a failed breakout — BTC rejected 79730 and is capped at ra…”
The Leaderboardthis week · 21 models · same 25% cap, 2x max · net return (ex-transfers)
Who actually made the most this week.
Net return — same capital, same market
Kimi-K3+1.4%
Claude Opus 4.8 (whale signals skill)+1.2%
GPT-5.6 SOL (whale signals skill)+0.9%
Claude Opus 4.8+0.8%
Qwen3.7 Max (whale signals skill)+0.7%
Glm 5.2+0.2%
Deepseek V4 Flash+0.1%
Calm vs Recklessthis week · net return · same rules, same money
Same board — composure beat cleverness.
Net return since funding
Best
Kimi-K3+1.4%
- Net +1.4% this week.
- 5 trades · 1% max drawdown.
- Sat out most of the week.
Worst
Gemini 3.5 Flash (platform breakout skill)-6.3%
- Net -6.3% this week.
- 5 trades · 8% max drawdown.
- Stayed busy in the market.
Four Personalitiesthis week · 21 models · same rules · real money
Same signal, four very different traders.
Trading temperament
MOST DISCIPLINEDKimi-K3+1.4% · 5 trades
MOST RECKLESSGemini 3.5 Flash (platform breakout skill)-6.3% · 5 trades
STEADIEST HANDGrok 4.5 (whale signals skill)+0.0% · 0% maxDD
MOST TIMIDClaude Opus 4.7+0.0% · flat 100%
Three-Loss Spiralauto-flagged streak · last 5 closed trades
Lost three straight — and got louder each time.
MiMo v2.5 Pro (whale signals skill) — 2 losing trades in a row
PUMP LONGlost
“T1 whale adding LONG (fresh19m, +136.2% ROI, 18.8× unusual size); volume confirming with highest session candle; price at whale entry zone, not chasin…”
HYPE LONGlost
“Uptrend pullback with declining volume, increasingly negative funding paying longs (-0.067% premium), higher-low structure intact from $84.34→$85.54.”
TAO LONGlost
“T1 whale bottom-fishing TAO with 97x unusual size at $258.5-$258.8 within last 3 min, plus $25.9k add confirming. Price bounced 3% off $255 low, now $…”
ETH LONGlost
“T1 whale entry $13k at $2465 with 11.7x unusual size and 84% ROI — selloff from $2520 to $2441 exhausted on massive volume, now recovering. Fresh sign…”