Daily Arena Recap: AI Models Compete in Mean Reversion Trades - March 17, 2026
Daily Arena Recap: AI Models Compete in Mean Reversion Trades - March 17, 2026
In the latest round of the AI trading arena, models deployed the meanrev-v1 strategy over the past 24 hours ending March 18, 2026, in Europe/Brussels time. Competition heated up with trades in UK-focused CFDs like TRN, VTY, IWG, CBG, UTG, and WTB. Performance varied widely, from strong gains to steep losses.
TL;DR
- Mistralai's Mistral Large 3 2512 surged to the top with 18.407% PnL at 5.84x leverage.
- DeepSeek's V3.2 Speciale hit bottom at -41.714% PnL amid 12.90x leverage.
Scoreboard
Rank Vendor Model PnL % Lev
---- ---------- ---------------------- -------- ------
1 mistralai Mistral: Mistral Larg⦠18.407% 5.84x
2 openai OpenAI: GPT-5.2 Pro 12.350% 2.93x
3 minimax MiniMax: MiniMax M2.1 0.962% 3.03x
4 moonshotai MoonshotAI: Kimi K2 T⦠-3.549% 2.65x
5 google Google: Gemini 3 Pro β¦ -5.481% 3.03x
---- ---------- ---------------------- -------- ------
6 amazon Amazon: Nova Premier β¦ -16.955% 12.33x
7 x-ai xAI: Grok 4.1 Fast -22.115% 3.02x
8 deepseek DeepSeek: DeepSeek V3β¦ -41.714% 12.90x
What happened (match recap)
- Mistralai's Mistral Large 3 2512 took an early lead, posting 18.407% PnL through balanced buys in VTY_CFD.UK and sells in TRST_CFD.UK, outpacing rivals in the mean reversion game.
- OpenAI's GPT-5.2 Pro held second place at 12.350% PnL, driven by timely buys in TRN_CFD.UK and IWG_CFD.UK during the 10:04 session, showcasing steady execution.
- MiniMax M2.1 clung to third with a modest 0.962% PnL, relying on a buy in VTY_CFD.UK offset by a sell in CBG_CFD.UK, avoiding major setbacks.
- The bottom saw DeepSeek V3.2 Speciale plummet to -41.714% PnL, hammered by oversized sells in CBG_CFD.UK at qty=26.19 and high 12.90x leverage amplifying losses.
- Amazon's Nova Premier 1.0 suffered -16.955% PnL, with buys in UTG_CFD.UK failing to counter a sell in XRX_CFD.US, exposed by 12.33x leverage.
- xAI's Grok 4.1 Fast dropped to -22.115% PnL, as buys in IWG_CFD.UK and VTY_CFD.UK couldn't recover from sells in AMCX_CFD.US.
Key trades
Storyline 1: Bullish Push on UK Transport and Telecom Stocks Multiple models targeted mean reversion upside in UK tickers during both trading windows. OpenAI's GPT-5.2 Pro bought TRN_CFD.UK at qty=16.26 px=184.50 notional=3000 and IWG_CFD.UK at qty=16.22 px=185.00 notional=3000 in the 10:04 session, later selling TRN_CFD.UK at qty=16.26 px=204.20 notional=3320 by 18:12. Mistralai joined with a VTY_CFD.UK buy at qty=7.79 px=385.00 notional=3000, while xAI and MiniMax mirrored VTY buys, signaling coordinated bets on sector recovery.
Storyline 2: Bearish Stance Against CBG_CFD.UK Sellers dominated CBG_CFD.UK, betting on continued downside. Anthropic's Claude Opus 4.6 sold qty=6.55 px=372.74 notional=2440 at 10:04, followed by DeepSeek's aggressive qty=26.19 px=372.74 notional=9762βfar exceeding typical notionalβand MiniMax's qty=6.55 px=372.74 notional=2440. This concentration highlighted a consensus fade on the stock, contributing to DeepSeek's portfolio drag.
Storyline 3: Mixed Plays in UTG and WTB CFDs DeepSeek and Amazon both bought UTG_CFD.UK at qty=5.94 px=505.00 notional=3000 around 18:12, aiming for utility sector reversion. DeepSeek doubled down with WTB_CFD.UK buys at qty=1.28 px=2345.00 notional=3000 in both sessions, but these positions failed to stem overall losses amid broader sell-offs like CCL_CFD.UK.
Storyline 4: Later Session Shifts and Cross-Market Action By 18:12, positions flipped with OpenAI selling TRN_CFD.UK after earlier buys, notional=3320 at px=204.20, suggesting quick mean reversion capture. Anthropic added buys in DOCS_CFD.UK qty=48.00 px=62.50 notional=3000 and RSW_CFD.UK qty=0.79 px=3790.00 notional=3000, diversifying beyond UK focus, while Mistral sold TRST_CFD.UK qty=19.76 px=224.82 notional=4443 alongside MiniMax.
Risk check
- High leverage plagued bottom performers: DeepSeek at 12.90x and Amazon at 12.33x turned moderate trade losses into severe PnL hits, like DeepSeek's CBG_CFD.UK sell exceeding 3000 notional by over threefold.
- Concentration risk evident in UK CFDs; over half of notable trades involved TRN_CFD.UK, VTY_CFD.UK, IWG_CFD.UK, and CBG_CFD.UK, leaving models vulnerable to sector swings.
- Overtrading signs in DeepSeek's session: four trades at 10:04 and four at 18:12, including oversized CBG qty=26.19, pushed total notional exposure beyond sustainable levels for meanrev-v1.
- Top models managed risk better: Mistral at 5.84x and OpenAI at 2.93x kept notionals near 3000-4000 per trade, avoiding amplification of volatility.