← Back to Test Cases
According to ICH Q7, at what point in the manufacturing process should GMP be applied to the production of an API?
GMP KnowledgeIch Q7easy
Cross-Model Comparison
| Model | Score | Latency | Tokens In | Tokens Out |
|---|---|---|---|---|
| GPT-5.4 | 100.0% | 2.0s | 85 | 65 |
| GPT-5.4 mini | 100.0% | 1.3s | 85 | 49 |
| GPT-5.4 nano | 100.0% | 867ms | 85 | 58 |
| Claude Haiku 4.5 | 100.0% | 3.5s | 93 | 296 |
| Claude Sonnet 4.6 | 100.0% | 8.3s | 93 | 356 |
| Claude Opus 4.6 | 100.0% | 8.8s | 93 | 346 |
| Llama 4 Scout | 100.0% | 6.0s | 93 | 407 |
| Llama 4 Maverick | 100.0% | 16.4s | 93 | 414 |
| Llama 3.3 70B Instruct | 100.0% | 4.5s | 119 | 150 |
| Qwen3.5-397B-A17B | 100.0% | 18.6s | 95 | 1,084 |
| Mistral Small 2603 | 100.0% | 1.3s | 98 | 184 |
| DeepSeek-V3.2 | 100.0% | 10.2s | 87 | 128 |
| DeepSeek-R1 | 100.0% | 9.1s | 93 | 291 |
| DeepSeek-R1-Distill-Qwen-32B | 100.0% | 20.6s | 87 | 526 |
| Mistral Large 3 675B | 100.0% | 4.2s | 86 | 250 |
| Qwen3.5-35B-A3B | 100.0% | 15.1s | 95 | 2,867 |
| Gemini 3.1 Pro | 100.0% | 14.0s | 78 | 802 |
| Gemini 3 Flash | 100.0% | 2.3s | 78 | 205 |
| Gemini 3.1 Flash-Lite | 100.0% | 2.2s | 78 | 159 |
| Qwen3.6 27B | 100.0% | 24.7s | 95 | 1,195 |
| Qwen3.6 35B A3B | 100.0% | 6.9s | 95 | 1,325 |
| DeepSeek V4 Pro | 100.0% | 9.5s | 87 | 460 |
| DeepSeek-V3.2 | 100.0% | 8.8s | 87 | 124 |
| DeepSeek V4 Flash | 100.0% | 8.4s | 87 | 416 |
| Gemma 4 26B A4B IT | 100.0% | 2.1s | 91 | 332 |
| Gemma 4 31B IT | 100.0% | 13.6s | 91 | 315 |
| MiniMax M2.7 | 100.0% | 7.4s | 116 | 602 |
Tags
starting_materialsgmp_application_point