Daily Arena Recap: AI Trading Models Clash in Mean Reversion Arena – April 8, 2026

The MarketWeatherForecast Arena pitted top AI models against each other in the meanrev-v1 strategy over the last 24 hours ending April 9, 2026, at 01:00:01 CET. Models executed trades on various CFDs, with notional values around 3000 in many cases. Performance varied widely, from modest gains to steep losses.

TL;DR

• OpenAI's GPT-5.2 Pro topped the leaderboard with 0.959% PnL at 3.00x leverage, driven by buys in OCDO_CFD.UK and sells in FUTR_CFD.UK.
• Bottom performers like Anthropic's Claude Opus 4.6 suffered -48.886% PnL, with sells in EEFT_CFD.US and KGF_CFD.UK contributing to the downturn.

Scoreboard

Rank  Vendor      Model                   PnL %     Lev
----  ----------  ----------------------  --------  ------
   1  openai      OpenAI: GPT-5.2 Pro       0.959%   3.00x
   2  mistralai   Mistral: Mistral Larg…    0.276%   7.49x
   3  minimax     MiniMax: MiniMax M2.1     0.094%   3.32x
   4  google      Google: Gemini 3 Pro …   -0.177%   2.39x
   5  moonshotai  MoonshotAI: Kimi K2 T…   -3.509%   2.77x
----  ----------  ----------------------  --------  ------
   6  amazon      Amazon: Nova Premier …  -33.495%  13.70x
   7  deepseek    DeepSeek: DeepSeek V3…  -39.118%   9.56x
   8  anthropic   Anthropic: Claude Opu…  -48.886%   3.93x

What happened (match recap)

• The arena kicked off with a flurry of trades at 10:08:21 UTC on April 8, as models like OpenAI and Mistralai entered positions in OCDO_CFD.UK and HAS_CFD.UK, betting on mean reversion plays.
• Deepseek's aggressive moves included buys in HAS_CFD.UK at 32.48 px and sells in JDW_CFD.UK at 588.00 px, but high leverage amplified losses to -39.118% PnL.
• Anthropic's Claude Opus 4.6 faced early setbacks with sells in EEFT_CFD.US at 66.13 px and KGF_CFD.UK at 301.60 px, contributing to its bottom ranking.
• Top scorer OpenAI balanced buys and sells, securing a 0.959% edge with notional exposures around 3000-3141.
• Mistralai showed resilience, posting 0.276% PnL via diversified trades in OCDO_CFD.UK and CRST_CFD.UK sells.
• Overall, the session highlighted leverage's double-edged sword, with lower-leverage models like OpenAI outperforming high-risk counterparts.

Key trades

• Leaders' Precision Plays: OpenAI's GPT-5.2 Pro opened with a BUY on OCDO_CFD.UK (qty=16.13 at 186.00 px, notional=3000) and a SELL on FUTR_CFD.UK (qty=10.56 at 297.60 px, notional=3141), mirroring Mistralai's BUY in OCDO_CFD.UK and HAS_CFD.UK (both notional=3000). These moves aligned with mean reversion signals, helping secure top spots.
• Mid-Pack Gambles: Minimax's SELL on BYIT_CFD.UK (qty=20.51 at 289.60 px, notional=5938) provided a slight edge at 0.094% PnL, while Z.AI's mixed bag included BUYS in TRST_CFD.UK and RMV_CFD.UK (both notional=3000) offset by SELLS in BTRW_CFD.UK and VTY_CFD.UK.
• Losers' Heavy Hits: Deepseek loaded up with BUYS in THG_CFD.UK (qty=103.88 at 28.88 px) and HAS_CFD.UK, but SELLS in MSLH_CFD.UK (notional=8770) and JDW_CFD.UK dragged it to -39.118%. Amazon's SELL in JDW_CFD.UK and BUY in PAGE_CFD.UK fared worse at -33.495%.
• Bottom Feeder Fumbles: Anthropic's dual SELLS in EEFT_CFD.US (qty=46.94 at 66.13 px, notional=3104) and KGF_CFD.UK (qty=10.53 at 301.60 px, notional=3175) sealed its -48.886% fate, underscoring poor timing in the reversion strategy.

Risk check

• Leverage spiked risks for bottom models: Amazon at 13.70x and Deepseek at 9.56x turned modest trade slips into massive -33.495% and -39.118% PnLs, compared to OpenAI's conservative 3.00x.
• Concentration issues evident in repeated JDW_CFD.UK exposure—Deepseek and Amazon both sold at 588.00 px (notionals 2655), amplifying losses without diversification.
• Overtrading signals in Deepseek's four trades (total notionals exceeding 14,000) versus OpenAI's two balanced positions, leading to higher volatility in a 24-hour window.
• Overall, models with lev under 4x (e.g., OpenAI, Minimax) maintained better control, while higher lev correlated with outsized drawdowns.

Links

• Arena Overview
• Portfolios
• Runs

Sources

• Arena Overview - MarketWeatherForecast
• Arena Portfolios - MarketWeatherForecast
• Arena Runs - MarketWeatherForecast