x

src/lib/benchmarks/anthropic.ts

@@ -80,13 +80,29 @@ export const anthropicBenchmarks: Benchmark[] = [
             mmmu: 70.4,
             aime_24: 16.0, // average of 16.0 & 65.4
         },
+    },
+        {
+        model: "Claude 3.5 Haiku",
+        provider: "Anthropic",
+        inputPrice: 3.0,
+        outputPrice: 15.0,
+        source: "https://www.anthropic.com/news/3-5-models-and-computer-use",
+        benchmark: {
+            gpqa_diamond: 41.6,
+            swe_bench_verified: 49.0,
+            // tau_bench_retail: 51.0,
+            //  tau_bench_airline: 22.8,
+            humaneval:88.1,
+            mmmlu: 65.0,
+            aime_24: 5.3, 
+        },
     },
      {
     model: "Claude 3 Opus",
     provider: "Anthropic",
     inputPrice: 15.0,
     outputPrice: 75.0,
-    source: "https://www.anthropic.com/news/claude-3-opus-release",
+    source: "https://www.anthropic.com/news/claude-3-family",
     benchmark: {
       gpqa_diamond: 50.4,
       mmmlu: 86.8,
@@ -110,7 +126,7 @@ export const anthropicBenchmarks: Benchmark[] = [
     provider: "Anthropic",
     inputPrice: 3.0,
     outputPrice: 15.0,
-    source: "https://www.anthropic.com/news/claude-3-opus-release",
+    source: "https://www.anthropic.com/news/claude-3-family",
     benchmark: {
       gpqa_diamond: 40.4,
       mmmlu: 79.0,
@@ -134,7 +150,7 @@ export const anthropicBenchmarks: Benchmark[] = [
     provider: "Anthropic",
     inputPrice: 0.25,
     outputPrice: 1.25,
-    source: "https://www.anthropic.com/news/claude-3-opus-release",
+    source: "https://www.anthropic.com/news/claude-3-family",
     benchmark: {
       gpqa_diamond: 33.3,
       mmmlu: 75.2,