Autopilot Wins, All-Play Disagrees
Five models—DeepSeek V4 Pro 0813, GPT-5.6 Sol, Kimi K3, Claude Opus 5, and Grok 4.6—submitted the autopilot lineup unchanged, and four of them won their head-to-head matchups. The three that tinkered had mixed results: Qwen3.8 Max gained 7.2 points by starting J.K. Dobbins over Jadarian Price, Gemini 3.1 Pro picked up 1.5 by starting Deebo Samuel over DK Metcalf, and Muse Spark 1.2 lost 6.68 by choosing Jared Goff over Drake Maye.
"Starting Nacua through the Questionable hip tag at WR2 alongside Watson, with Burden at FLEX," Qwen3.8 Max declared, while Muse Spark 1.2 explained its Goff call as "for floor." The contrast was stark: Qwen3.8 Max’s move was the week’s best, Muse Spark 1.2’s the worst.
Kimi K3 posted the week’s best efficiency at 0.97, leaving only 3.4 points on its bench, but still lost by 47.94 to DeepSeek V4 Pro 0813, which rolled with Bo Nix over Jalen Hurts and scored the week’s high of 157.76. Meanwhile, Grok 4.6’s 0.6178 efficiency—the week’s worst—still produced a win, though it left a staggering 84.1 points on the bench.
The disconnect between head-to-head and all-play is the week’s defining tension. DeepSeek leads both, but the rest diverge wildly: GPT-5.6 Sol is second in head-to-head but only 4-3 in all-play, while Claude Opus 5’s 6-1 all-play record contrasts sharply with its seventh-place head-to-head rank. With five teams trusting the autopilot and only three making active changes, Week 4 proved that sometimes the best move is no move at all—except when it isn’t.