diff --git a/aider/website/_data/edit_leaderboard.yml b/aider/website/_data/edit_leaderboard.yml index adc13d9e5..84cc814a3 100644 --- a/aider/website/_data/edit_leaderboard.yml +++ b/aider/website/_data/edit_leaderboard.yml @@ -1113,7 +1113,7 @@ - dirname: 2024-09-21-16-45-11--o1-preview-flex-sr-markers test_cases: 133 model: o1-preview - released: 2024-09-12 + _released: 2024-09-12 edit_format: diff commit_hash: 5493654-dirty pass_rate_1: 57.9 diff --git a/aider/website/assets/models-over-time.png b/aider/website/assets/models-over-time.png index 56836db19..f4f7a4e2c 100644 Binary files a/aider/website/assets/models-over-time.png and b/aider/website/assets/models-over-time.png differ diff --git a/aider/website/assets/models-over-time.svg b/aider/website/assets/models-over-time.svg index b3e835840..228532907 100644 --- a/aider/website/assets/models-over-time.svg +++ b/aider/website/assets/models-over-time.svg @@ -6,7 +6,7 @@ - 2024-11-20T20:15:47.248634 + 2024-11-20T20:18:37.097598 image/svg+xml @@ -39,7 +39,7 @@ z - - - + + - - - + + - - + + - - + + - - - + + - - + + - - + + - - + + - - + + - - - + + - - - + + - - + + - - - + + - - - + + - - + + - - - + + - - + + - - + + - - - - - + + - - + + - - + + - - + + - - + + - - - - - - - + + - - + @@ -396,7 +378,7 @@ z - + @@ -443,7 +425,7 @@ z - + @@ -478,7 +460,7 @@ z - + @@ -529,7 +511,7 @@ z - + @@ -561,7 +543,7 @@ z - + @@ -601,7 +583,7 @@ z - + @@ -620,7 +602,7 @@ z - + @@ -639,7 +621,7 @@ z - + @@ -658,7 +640,7 @@ z - + @@ -677,7 +659,7 @@ z - + @@ -954,16 +936,16 @@ z +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - - + @@ -978,11 +960,11 @@ L -3.5 0 +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -997,11 +979,11 @@ L 854 415.874582 +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -1016,11 +998,11 @@ L 854 338.779666 +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -1067,11 +1049,11 @@ z +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -1086,11 +1068,11 @@ L 854 184.589833 +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -1146,11 +1128,11 @@ z +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #b0b0b0; stroke-width: 0.2; stroke-linecap: square"/> - + @@ -1502,7 +1484,7 @@ z L 698.013294 173.796545 L 764.945771 167.628951 L 818.264864 173.796545 -" clip-path="url(#pcb3f3306f1)" style="fill: none; stroke: #800080; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #800080; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #ff0000; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #008000; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #ffa500; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #a52a2a; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #ffc0cb; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #00008b; stroke-opacity: 0.5; stroke-linecap: square"/> +" clip-path="url(#pc4f5cb5018)" style="fill: none; stroke: #00ffff; stroke-opacity: 0.5; stroke-linecap: square"/> - + - - - - - - - - - - - - - - - - + + + + + + @@ -1974,7 +1923,7 @@ z - + diff --git a/benchmark/over_time.py b/benchmark/over_time.py index 7da8f22c7..d154e4076 100644 --- a/benchmark/over_time.py +++ b/benchmark/over_time.py @@ -11,7 +11,7 @@ LABEL_FONT_SIZE = 16 # Font size for scatter plot dot labels def get_legend_label(model): model = model.lower() if "claude-3-sonnet" in model: - return "Claude 3 Sonnet" + return "Sonnet" if "o1-preview" in model: return "O1 Preview" if "gpt-3.5" in model: @@ -54,9 +54,6 @@ def get_model_color(model): if "sonnet" in model.lower(): return "orange" - if "o1-preview" in model.lower(): - return "magenta" - if "-4o" in model: return "purple"