293 models ranked for ML engineering. Scored with bonuses for reasoning (architecture decisions), large context (reading full codebases), large output (complete implementations), JSON mode, and function calling.
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 ProOpenAI | 91 |
| 2 | GPT-5.2 ProOpenAI | 90 |
| 3 | GPT-5 ProOpenAI | 90 |
| 4 | o3 ProOpenAI | 82 |
| 5 | Claude Opus 4.1Anthropic | 81 |
| 6 | o1-proOpenAI | 77 |
| 7 | o3 Deep ResearchOpenAI | 74 |
| 8 | Claude Opus 4Anthropic | 76 |
| 9 | Claude Opus 4.6Anthropic | 71 |
| 10 | Claude Opus 4.5Anthropic | 70 |
| 11 | GPT-5.4OpenAI | 70 |
| 12 | Claude Sonnet 4.5Anthropic | 69 |
| 13 | Qwen3 VL 30B A3B ThinkingAlibaba | 69 |
| 14 | Qwen3 VL 235B A22B ThinkingAlibaba | 69 |
| 15 | GPT-5.2OpenAI | 68 |
| 16 | Gemini 3.1 Pro Preview Custom ToolsGoogle | 68 |
| 17 | Gemini 3.1 Pro PreviewGoogle | 68 |
| 18 | Gemini 3 Pro PreviewGoogle | 68 |
| 19 | Claude Sonnet 4.6Anthropic | 68 |
| 20 | GPT-5.1OpenAI | 67 |
| 21 | GPT-5.3-CodexOpenAI | 67 |
| 22 | GPT-5.2-CodexOpenAI | 67 |
| 23 | GPT-5OpenAI | 67 |
| 24 | Gemini 3 Flash PreviewGoogle | 66 |
| 25 | o4 Mini Deep ResearchOpenAI | 66 |
| 26 | GPT-5.1-Codex-MaxOpenAI | 66 |
| 27 | Gemini 3.1 Flash Lite PreviewGoogle | 66 |
| 28 | Gemini 2.5 ProGoogle | 66 |
| 29 | Gemini 2.5 Flash Lite Preview 09-2025Google | 65 |
| 30 | GPT-5 MiniOpenAI | 65 |
Design neural network architectures, select hyperparameters, and choose training strategies. Reasoning models analyze trade-offs between model complexity and performance.
Generate PyTorch, TensorFlow, and scikit-learn code for training pipelines, data loaders, custom layers, and evaluation scripts. Large output produces complete implementations.
Analyze experiment results, suggest next steps, and document findings. JSON mode structures experiment metadata for tools like MLflow, W&B, and Neptune.
Create model serving configs, write Docker/Kubernetes manifests for inference, and build monitoring dashboards. Function calling integrates with deployment APIs.