release 3 and strawberry

all_outputs.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1e71760577ccfff18dd80b4c60bed272104cef4bf57e904da6203108840320f7
-size 169602473
+oid sha256:14e11aaec8c6bbb3420b7ece68bad0f7c4d354da886defa2ad35d229232e1bc2
+size 238813493

app.py

@@ -103,15 +103,18 @@ def problem(problem_idx):
 
 
 mini_models = [
-    "DeepSeek-V2",
+    # "DeepSeek-V2",
     "DeepSeekCoder-V2",
-    "DSCoder-33b-Ins",
-    "LLama3-70b-Ins",
-    "GPT-4-Turbo-2024-04-09",
+    # "DSCoder-33b-Ins",
+    "LLama3.1-70b-Ins",
+    "LLama3.1-405b-Ins-FP8",
+    # "GPT-4-Turbo-2024-04-09",
     "GPT-4O-2024-05-13",
     "Claude-3-Opus",
-    "Claude-3-Sonnet",
-    "Gemini-Pro-1.5-May",
+    # "Claude-3-Sonnet",
+    "Gemini-Pro-1.5-August",
+    "O1-Mini (N=1)",
+    "O1-Preview (N=1)",
 ]
 
 

problems.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d91ec0f4e3e94ecf089a7725086b8d40560b388f0a69820c6acd9d96a082860e
-size 909748
+oid sha256:3e1845480559405c5a8a3d1d17f5d0389a72684eb3ce86a602cf82b14d97a84a
+size 1065970