Results as of 2026-09-16T04:34:41Z: 12 models, 134,316 answers

README.md

@@ -39,17 +39,17 @@ Up and to the left is better. The dashed line is the cost-performance frontier w
 
 | #             | Model | TOTAL | Sweep cost |
 | ------------- | ----- | ----- | ---------- |
-| 1 | GPT-6 Astra | 83.60 | $542.16 | $111.50 |
-| 2 | Gemini 3.8 Flash | 75.93 | $16.93 | $0.70 |
-| 3 | GPT-5.6 Sol | 75.56 | $106.31 | $22.28 |
-| 4 | Qwen 3.8 Max | 75.38 | $89.26 | $16.11 |
-| 5 | Muse Spark 1.3 | 73.10 | $64.03 | $3.57 |
-| 6 | GPT-5.6 Luna | 65.29 | $10.69 | $2.23 |
-| 7 | MiniMax M3 | 64.35 | $11.27 | $0.57 |
-| 8 | Gemini 2.5 Flash Lite | 64.04 | $5.53 | $0.44 |
-| 9 | Qwen 3.6 35B A3B | 62.96 | $2.67 | $0.34 |
-| 10 | Qwen 3.8 27B | 61.64 | $5.56 | $0.58 |
-| 11 | Gemma 4 31B | 59.66 | $0.49 | $0.03 |
+| 1 | GPT-6 Astra | 83.60 | $542.16 | $0.02916 | $0.00113 | $105.47 |
+| 2 | Gemini 3.8 Flash | 75.93 | $16.93 | $0.00006 | $0.00077 | $0.58 |
+| 3 | GPT-5.6 Sol | 75.56 | $106.31 | $0.00579 | $0.00007 | $20.88 |
+| 4 | Qwen 3.8 Max | 75.38 | $89.26 | $0.00373 | $0.00183 | $14.25 |
+| 5 | Muse Spark 1.3 | 73.10 | $64.03 | $0.00017 | $0.00349 | $2.18 |
+| 6 | GPT-5.6 Luna | 65.29 | $10.69 | $0.00058 | $0.00001 | $2.10 |
+| 7 | MiniMax M3 | 64.35 | $11.27 | $0.00009 | $0.00058 | $0.57 |
+| 8 | Gemini 2.5 Flash Lite | 64.04 | $5.53 | $0.00003 | $0.00032 | $0.24 |
+| 9 | Qwen 3.6 35B A3B | 62.96 | $2.67 | $0.00012 | $0.00000 | $0.43 |
+| 10 | Qwen 3.8 27B | 61.64 | $5.56 | $0.00020 | $0.00003 | $0.75 |
+| 11 | Gemma 4 31B | 59.66 | $0.49 | $0.00001 | $0.00000 | $0.05 |
 
 
 Also in the data but left out of the figure and this table: Reka Edge (39.78, 7B edge model scoring 39.8; kept in the data, left out of the headline figures as an outlier). `models.featured` marks them.

models.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd17d5b85f2c285bb0ea020f803b8a6041ec8a4c760ce69bde8e8824c165ea29
-size 9876
+oid sha256:817e2068ee96e180fb3851ffa30de0901d43253f8421b3efdd866835ba5c325c
+size 10354