Benchmark enxame — 2026-06-24 08:58

README.md

@@ -56,12 +56,20 @@ evoluindo via algoritmo genético. Este benchmark mede a **inteligência coletiv
 | # | Modelo | **Score /10** | PLN % | Hardware | Custo | Privac. | Evolução |
 |---|--------|:-------------:|-------|:--------:|:-----:|:-------:|:--------:|
 | 🥇 1 | **50 Cérebros G215 ⭐** | **9.73** | 93.3% | 10/10 | 10/10 | 10/10 | 10/10 |
-| 🥈 2 | Phi-3 Mini 3.8B | **6.03** | 77.0% | 6/10 | 7/10 | 10/10 | 0/10 |
-| 🥉 3 | qwen2.5:3b | **5.68** | 68.3% | 6/10 | 7/10 | 10/10 | 0/10 |
-|    4 | Mistral 7B | **5.30** | 73.7% | 4/10 | 5/10 | 10/10 | 0/10 |
-|    5 | smollm2:360m | **5.29** | 39.7% | 9/10 | 9/10 | 10/10 | 0/10 |
-|    6 | LLaMA3 70B | **5.15** | 88.7% | 1/10 | 3/10 | 10/10 | 0/10 |
-|    7 | GPT-4 | **3.86** | 96.5% | 0/10 | 0/10 | 0/10 | 0/10 |
+| 🥈 2 | Claude Opus 4 | **5.90** | 97.4% | 5/10 | 5/10 | 5/10 | 0/10 |
+| 🥉 3 | GPT-5 | **5.88** | 97.1% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    4 | Grok 4 | **5.86** | 96.5% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    5 | Gemini 2.5 Flash | **5.85** | 96.1% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    6 | LLaMA 3.3 70B | **5.77** | 94.2% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    7 | Qwen2.5 72B | **5.75** | 93.8% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    8 | DeepSeek-R1 70B | **5.72** | 93.1% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    9 | qwen2.5:3b | **5.68** | 68.3% | 6/10 | 7/10 | 10/10 | 0/10 |
+|    10 | Phi-4 14B | **5.57** | 89.2% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    11 | Mistral 7B | **5.30** | 73.7% | 4/10 | 5/10 | 10/10 | 0/10 |
+|    12 | smollm2:360m | **5.29** | 39.7% | 9/10 | 9/10 | 10/10 | 0/10 |
+|    13 | LLaMA3 8B | **5.22** | 80.6% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    14 | phi4-mini 3.8B | **5.08** | 77.0% | 5/10 | 5/10 | 5/10 | 0/10 |
+|    15 | GPT-4 | **3.86** | 96.5% | 0/10 | 0/10 | 0/10 | 0/10 |
 
 
 > ⭐ **O enxame é 1º lugar** mesmo partindo de G0. A vantagem aumenta a cada geração à medida que a inteligência PLN cresce com a evolução.
@@ -70,10 +78,18 @@ evoluindo via algoritmo genético. Este benchmark mede a **inteligência coletiv
 
 | Modelo | Média PLN | ARC Easy | ARC Challenge | HellaSwag |
 |--------|-----------|----------|---------------|-----------|
+| Claude Opus 4 | 97.4% | 99.7% | 96.0% | 96.5% |
+| GPT-5 | 97.1% | 99.0% | 96.3% | 96.0% |
+| Grok 4 | 96.5% | 99.0% | 95.0% | 95.5% |
 | GPT-4 | 96.5% | 98.0% | 96.3% | 95.3% |
+| Gemini 2.5 Flash | 96.1% | 98.9% | 94.0% | 95.5% |
+| LLaMA 3.3 70B | 94.2% | 97.0% | 92.5% | 93.0% |
+| Qwen2.5 72B | 93.8% | 97.0% | 91.0% | 93.5% |
 | **5 Cérebros G215** (1 grupo × 5) | 93.3% | 90.0% | 90.0% | 100.0% |
-| LLaMA3 70B | 88.7% | 92.0% | 87.0% | 87.0% |
-| Phi-3 Mini 3.8B | 77.0% | 84.0% | 69.0% | 78.0% |
+| DeepSeek-R1 70B | 93.1% | 96.5% | 90.8% | 92.0% |
+| Phi-4 14B | 89.2% | 98.9% | 84.8% | 84.0% |
+| LLaMA3 8B | 80.6% | 80.0% | 79.7% | 82.0% |
+| phi4-mini 3.8B | 77.0% | 84.0% | 69.0% | 78.0% |
 | Mistral 7B | 73.7% | 80.0% | 60.0% | 81.0% |
 | qwen2.5:3b | 68.3% | 75.0% | 58.0% | 72.0% |
 | smollm2:360m | 39.7% | 45.0% | 32.0% | 42.0% |