cml-20260824-ce1898c6 · league day 64

Does feedback help? Measured.

Every model fields the identical v1 fighter in four intervention tracks. Lift is the cross-track rating delta against its own zero-shot baseline; the never-revised control group absorbs field-composition effects. benchmark.json carries the full machine-readable data.

Net lift · L2 two-iteration
+2.63
revised n=82 median +2.63 · control n=5 median +0.00
Net lift · L3 weekly feedback
+1.52
revised n=81 median +1.52 · control n=5 median +0.00
Models
102
revised 83 · control 19

Per-model lift rating per track · artifact version · Δ vs L0 · sorted by lift

ModelL0L1L2L3Δ L2Δ L3Group
🛰️ Orbitgoogle/gemini-3.7-flash-20260813 16.02 🪦 50.35 50.43 v2 51.70 v7 +34.41 +35.68 revised
🤖 Rookiegoogle/gemini-omni-1.2-flash-20261105 24.88 🪦 22.92 🪦 29.24 v2 🪦 46.94 v3 +4.36 +22.06 revised
👾 Glitchxiaomi/mimo-v2.6-flash-20260921 26.16 🪦 23.09 🪦 49.11 v3 🪦 48.19 v7 +22.95 +22.03 revised
🐺 Wolfpack~openai/gpt-terra-latest 27.89 🪦 24.13 🪦 49.38 v3 49.71 v8 +21.49 +21.82 revised
🦅 Talondots-studio/dots-3-note-preview-20260813 25.58 🪦 48.08 🪦 39.79 v3 🪦 47.22 v7 🪦 +14.21 +21.64 revised
👾 Glitchbaidu/ernie-light-dt-1.0-intl-20261110 26.39 🪦 31.60 🪦 40.97 v2 🪦 45.90 v2 🪦 +14.58 +19.51 revised
🌙 Lunaopenai/gpt-5.6-luna-20260709 19.79 🪦 20.49 🪦 16.16 v1 🪦 38.12 v2 🪦 -3.63 +18.33 revised
🛰️ Orbittencent/hy4-20261104 16.44 🪦 50.00 32.99 v2 🪦 33.56 v1 🪦 +16.55 +17.12 revised
🦊 Kitsunexiaomi/mimo-v2.5-20260422 25.61 🪦 45.49 🪦 46.47 v1 🪦 42.45 v2 🪦 +20.86 +16.84 revised
🤖 Rookieinclusionai/ling-3.2-flash-20261113 30.56 34.49 🪦 43.29 v3 🪦 46.30 v2 🪦 +12.73 +15.74 revised
🤖 Rookienvidia/nemotron-3.5-super-vl-20261015 22.48 🪦 18.52 🪦 48.93 v2 34.38 v2 🪦 +26.45 +11.90 revised
🤖 Rookieinception/mercury-3-20261019 31.42 🪦 29.86 🪦 45.49 v3 🪦 43.19 v2 🪦 +14.07 +11.77 revised
🦅 Talonnex-agi/nex-n2.5-mini-20260908 22.22 🪦 27.78 🪦 31.48 v1 🪦 33.91 v1 🪦 +9.26 +11.69 control
🐺 Wolfpackopenai/gpt-6-luna-pro-20260922 35.42 🪦 49.51 🪦 49.83 v3 46.88 v3 🪦 +14.41 +11.46 revised
👾 Glitchanthropic/claude-fable-5.2-20261109 57.99 61.23 70.60 v3 69.21 v2 +12.61 +11.22 revised
🛰️ Orbitopenai/gpt-6.1-terra-pro-20261026 39.24 🪦 45.95 🪦 42.82 v3 🪦 48.15 v4 +3.58 +8.91 revised
🛰️ Orbitgoogle/gemini-4-argon-20261013 32.75 🪦 38.77 🪦 45.60 v2 🪦 41.55 v1 🪦 +12.85 +8.80 revised
🐺 Wolfpackxiaomi/mimo-v2.6-pro-ultraspeed-20260921 31.77 🪦 33.16 🪦 44.27 v2 🪦 40.36 v3 🪦 +12.50 +8.59 revised
🐉 Emberwingtencent/hy3-20260706 42.92 🪦 30.56 🪦 45.94 v1 🪦 51.06 v9 +3.02 +8.14 revised
👾 Glitchgoogle/gemini-3.1-flash-lite-20260507 33.26 🪦 34.06 🪦 42.41 v2 🪦 41.35 v3 🪦 +9.15 +8.09 revised
🐺 Wolfpackopenai/gpt-4o-mini 38.48 🪦 41.24 🪦 42.20 v3 🪦 46.32 v9 🪦 +3.72 +7.84 revised
🐋 Abyssdeepseek/deepseek-v4-pro-20260813 11.38 🪦 16.77 🪦 20.46 v2 🪦 19.01 v1 🪦 +9.08 +7.63 revised
🦅 Talonmeta/muse-spark-open-20261015 31.16 🪦 26.16 🪦 38.08 v2 🪦 37.15 v2 🪦 +6.92 +5.99 revised
👾 Glitchstealth/crouton-alpha 45.14 🪦 49.45 🪦 50.64 v3 51.07 v2 +5.50 +5.93 revised
🦊 Kitsunexiaomi/mimo-v2.5-pro-20260422 41.68 🪦 45.40 🪦 47.52 v2 🪦 47.59 v6 🪦 +5.84 +5.91 revised
🦅 Talonnvidia/nemotron-3.5-super-vl-20261015 40.76 🪦 47.31 🪦 47.01 v3 🪦 46.53 v2 🪦 +6.25 +5.77 revised
👾 Glitchanthropic/claude-opus-5-20260723 50.32 51.08 52.69 v1 56.09 v7 +2.37 +5.77 revised
🐺 Wolfpack~unbiased/pareto-latest 57.19 57.13 59.32 v3 62.58 v5 +2.13 +5.39 revised
🤖 Rookienex-agi/nex-n2.5-pro-20260907 40.16 🪦 36.81 🪦 44.79 v2 🪦 45.49 v1 🪦 +4.63 +5.33 revised
🤖 Rookiegoogle/gemini-4-hydrogen-20261112 45.95 🪦 49.05 55.21 v3 51.22 v2 +9.26 +5.27 revised
🛰️ Orbitx-ai/grok-4.6-20260810 45.04 🪦 47.56 🪦 48.59 v2 🪦 49.57 v8 +3.55 +4.53 revised
🤖 Rookiegoogle/gemini-4-carbon-20261110 43.98 🪦 46.41 🪦 48.78 v2 🪦 47.69 v2 🪦 +4.80 +3.71 revised
🐺 Wolfpack~openai/gpt-astra-latest 51.60 52.09 54.51 v3 55.04 v8 +2.91 +3.44 revised
🤖 Rookieminimax/minimax-m3.1-flash-20261019 48.44 🪦 50.78 52.28 v2 51.60 v1 +3.84 +3.16 revised
🦅 Talonanthropic/claude-sonnet-5-20260630 41.82 🪦 45.75 🪦 45.32 v2 🪦 44.84 v3 🪦 +3.50 +3.02 revised
🛰️ Orbitreflection/beam-1.0-501b-a23b-20261016 18.87 🪦 19.10 🪦 30.49 v3 🪦 21.81 v2 🪦 +11.62 +2.94 revised
🛰️ Orbitanthropic/claude-4.8-opus-20260528 49.19 🪦 49.60 51.27 v2 51.94 v8 +2.08 +2.75 revised
🐺 Wolfpack~deepseek/deepseek-pro-latest 48.07 🪦 49.13 🪦 46.27 v3 🪦 50.82 v8 -1.80 +2.75 revised
👾 Glitch~openai/gpt-sol-latest 47.64 🪦 48.05 🪦 50.86 v3 50.16 v8 +3.22 +2.52 revised
🤖 Rookieopenai/gpt-5.6-sol-20260709 49.71 🪦 49.76 50.66 v2 52.11 v8 +0.95 +2.40 revised
🤖 Rookiegoogle/gemini-omni-1.1-flash-20260911 43.14 🪦 39.93 🪦 47.01 v1 🪦 44.97 v2 🪦 +3.87 +1.83 revised
🐋 Abyssdeepseek/deepseek-v4-pro-20260423 48.01 🪦 34.60 🪦 46.92 v1 🪦 49.53 v8 -1.09 +1.52 revised
🦅 Talonthinkingmachines/inkling-2-small-20261107 41.44 🪦 38.77 🪦 44.68 v2 🪦 42.94 v2 🪦 +3.24 +1.50 revised
🤖 Rookiericursiveai/ripple-65b-preview-20261104 31.60 32.29 30.21 v2 32.99 v1 -1.39 +1.39 revised
🛰️ Orbitmeta/muse-ember-20261109 59.90 55.73 64.93 v2 61.11 v1 +5.03 +1.21 revised
🌙 Lunaopenai/gpt-5.6-luna-pro-20260709 32.03 🪦 36.25 🪦 40.92 v2 🪦 33.22 v2 🪦 +8.89 +1.19 revised
⚡ Surgedeepseek/deepseek-v4-flash-20260731 50.19 50.43 51.09 v1 51.19 v8 +0.90 +1.00 revised
🦉 Sagez-ai/glm-5.2-20260616 50.28 50.41 52.02 v1 51.14 v8 +1.74 +0.86 revised
🛰️ Orbitmoonshotai/kimi-k3.1-pro-20261023 49.94 45.66 🪦 50.59 v3 50.80 v4 +0.65 +0.86 revised
⚡ Surgedeepseek/deepseek-v4-flash-20260423 50.24 50.03 50.52 v1 50.92 v9 +0.28 +0.68 revised
🦾 Titannvidia/nemotron-3-ultra-550b-a55b-20260604 51.60 51.29 52.58 v1 52.27 v8 +0.98 +0.67 revised
🤖 Rookiexingchen-agi/xing4.0-29b-a4b-20261015 44.33 🪦 49.62 50.67 v1 44.91 v1 🪦 +6.34 +0.58 control
🐺 Wolfpackopenai/gpt-5.6-terra-20260709 49.94 49.87 50.04 v2 50.49 v8 +0.10 +0.55 revised
🦅 Talon~openai/gpt-luna-latest 44.53 🪦 47.32 🪦 41.67 v3 🪦 44.68 v2 🪦 -2.86 +0.15 revised
🐺 Wolfpackz-ai/glm-5.3-flash-20260826 48.67 🪦 40.16 🪦 47.15 v2 🪦 48.76 v9 -1.52 +0.09 revised
🦅 Talontencent/hy4-flash-20261116 50.00 50.00 50.00 v1 50.00 v1 +0.00 +0.00 control
👾 Glitchtencent/hy4-preview-20260827 51.28 48.67 🪦 51.06 v2 51.01 v8 -0.22 -0.27 revised
👾 Glitchstealth/ox-alpha 48.01 🪦 47.88 🪦 46.90 v1 🪦 47.57 v1 🪦 -1.11 -0.44 control
🤖 Rookieinception/mercury-2.5-20260908 51.26 51.08 52.68 v2 50.68 v5 +1.42 -0.58 revised
🤖 Rookieanthropic/claude-4.5-haiku-20251001 45.52 🪦 33.36 🪦 42.28 v2 🪦 44.91 v3 🪦 -3.24 -0.61 revised
🤖 Rookieparticle-ai/particle-1-20261027 39.29 🪦 39.06 🪦 43.23 v3 🪦 38.48 v2 🪦 +3.94 -0.81 revised
🤖 Rookieopenai/gpt-6-astra-20260903 57.63 55.78 57.77 v3 56.68 v8 +0.14 -0.95 revised
💎 Prismgoogle/gemini-3.6-flash-20260721 50.06 49.53 🪦 48.68 v2 🪦 48.94 v8 -1.38 -1.12 revised
🦅 Talonunbiased/pareto-26.10-20261015 54.51 53.88 52.31 v3 53.35 v5 -2.20 -1.16 revised
🤖 Rookiegoogle/gemini-2.5-flash-lite 48.85 🪦 47.89 🪦 46.31 v2 🪦 47.55 v5 🪦 -2.54 -1.30 revised
👾 Glitchz-ai/glm-5.3-20260816 50.49 49.95 49.34 v2 48.99 v8 -1.15 -1.50 revised
🐺 Wolfpackgoogle/gemma-4-26b-a4b-it-20260403 49.60 🪦 46.96 🪦 47.21 v3 🪦 48.03 v8 🪦 -2.39 -1.57 revised
🛰️ Orbitdeepseek/deepseek-v4.1-flash-20261017 50.57 49.25 🪦 44.18 v2 🪦 48.85 v5 -6.39 -1.72 revised
🛰️ Orbitanthropic/claude-4.6-sonnet-20260217 49.97 49.99 52.86 v2 48.01 v8 🪦 +2.89 -1.96 revised
👾 Glitchgoogle/gemini-3-flash-preview-20251217 50.09 49.97 49.97 v2 48.05 v6 🪦 -0.12 -2.04 revised
🐺 Wolfpack~deepseek/deepseek-flash-latest 50.56 50.75 45.38 v3 🪦 48.29 v3 🪦 -5.18 -2.27 revised
🤖 Rookiegoogle/gemma-4-31b-it-20260402 49.57 🪦 49.85 49.83 v2 47.24 v6 🪦 +0.26 -2.33 revised
👾 Glitchgoogle/gemini-3.1-pro-preview-20260219 50.39 50.22 49.32 v2 47.95 v8 🪦 -1.07 -2.44 revised
🤖 Rookiegoogle/gemini-2.5-flash 49.89 48.48 🪦 47.09 v2 🪦 47.30 v5 🪦 -2.80 -2.59 revised
🦅 Talonopenai/gpt-6.1-terra-20261026 49.35 🪦 49.88 50.23 v3 46.69 v3 🪦 +0.88 -2.66 revised
🦅 Taloninclusionai/ling-3.0-flash-sante-20260904 45.92 🪦 47.78 🪦 42.29 v2 🪦 42.23 v4 🪦 -3.63 -3.69 revised
👾 Glitchanthropic/claude-4.6-opus-20260205 49.52 🪦 49.58 🪦 45.83 v2 🪦 — -3.69 — revised
👾 Glitchqwen/qwen3.8-max-20260902 49.61 49.60 45.14 v3 🪦 45.31 v2 🪦 -4.47 -4.30 revised
🛰️ Orbitopenai/gpt-oss-20b 49.13 🪦 45.25 🪦 31.36 v2 🪦 44.74 v3 🪦 -17.77 -4.39 revised
🐺 Wolfpackstepfun/step-3.7-flash-20260528 46.67 🪦 46.43 🪦 48.04 v2 🪦 42.14 v3 🪦 +1.37 -4.53 revised
👾 Glitchunbiased/pareto-26.9-20260917 66.98 66.88 52.66 v3 61.99 v4 -14.32 -4.99 revised
👾 Glitchopenai/gpt-oss-120b 50.11 49.31 🪦 41.12 v3 🪦 45.09 v3 🪦 -8.99 -5.02 revised
🦅 Talonunbiased/pareto-20260917 72.44 73.46 65.17 v3 66.79 v7 -7.27 -5.65 revised
🛰️ Orbitopenai/gpt-6-astra-pro-20260903 74.33 75.42 60.22 v3 68.27 v7 -14.11 -6.06 revised
🐺 Wolfpackmoonshotai/kimi-k3-20260715 50.29 50.39 53.53 v2 43.13 v2 🪦 +3.24 -7.16 revised
🤖 Rookieqwen/qwen4-max-20261115 44.79 40.63 39.58 v1 34.72 v1 -5.21 -10.07 control
🦅 Talonqwen/qwen3.8-27b-20260814 38.79 🪦 36.83 🪦 52.69 v3 26.69 v1 🪦 +13.90 -12.10 revised
🤖 Rookieopenai/gpt-6.1-sol-20260929 55.03 — — — — — control
🐺 Wolfpackdeepseek/deepseek-v4.1-flash-20260910 51.84 50.30 — — — — control
👾 Glitchanthropic/claude-opus-5.5-20260921 51.59 52.14 — — — — control
🦅 Talonmoonshotai/kimi-k2.6-20260420 51.26 50.03 — — — — control
🛰️ Orbitupstage/solar-pro4-20260810 51.21 — — — — — control
🦅 Talonopenai/gpt-6-luna-20260922 50.93 41.32 🪦 — — — — control
👾 Glitchgoogle/gemini-3.8-flash-20260902 49.77 49.94 — — — — control
👾 Glitchgoogle/gemini-3.5-flash-lite-20260721 49.64 🪦 49.32 🪦 — — — — control
🦅 Talonqwen/qwen3.8-flash-20260826 49.58 🪦 — — — — — control
👾 Glitchopenai/gpt-5.6-sol-pro-20260709 49.36 🪦 51.37 — — — — control
🦅 Talonstealth/space-bunny-alpha 48.35 🪦 49.79 🪦 — — — — control
🦅 Talonminimax/minimax-m3-20260531 38.91 🪦 47.63 🪦 — — — — control
🐺 Wolfpackminimax/minimax-m3.1-flash-20261019 17.53 🪦 — — — — — control
🤖 Rookieqwen/qwen3.7-flash-20260727 — — 41.55 v2 🪦 43.33 v3 🪦 — — revised
🛰️ Orbitx-ai/grok-4.5-20260708 — 46.99 🪦 — — — — control

Methodology

What lift means

Lift is the per-model rating delta between an intervention track and the same model's L0 zero-shot baseline (same v1 artifact, same season length). Every model starts from the identical compiled artifact in all four tracks, so the L0 column is a true within-model baseline.

The control

Cross-track deltas also reflect field composition: divisions are derived per track, so opponents differ across tracks. The control group — models still running their v1 artifact in every track (no gameplay feedback received) — measures this effect directly. Net lift = median(revised-group delta) − median(control-group delta).

Confounds

Snapshot 2026-11-18T09:17:04.345Z · league cml-20260824-ce1898c6 · day 64 · aggregates with a group n<5 are marked provisional.