Which AI models can you automate on Zapier? (GPT-5.6 Sol, Gemini 3.7 Flash, Opus 5, and more)
Zapier has published a benchmark-driven guide to which AI models can be automated in Zapier workflows, including OpenAI’s GPT-5.6 family, Anthropic’s Opus 5, and Google’s Gemini 3.7 Flash.
Zapier introduced AutomationBench, a proprietary benchmark that evaluates AI models on multi-step business workflows rather than static prompts. The tool measures how often models complete complex, realistic tasks such as resolving scheduling conflicts, reconciling data across systems, or enforcing business rules. Scores reflect only whether the task is completed correctly, not the method used, prioritizing reliability over cost or speed.
OpenAI’s GPT-5.6 family—Sol, Terra, and Luna—are now available on Zapier, each optimized for different use cases. Sol handles high-stakes, one-shot workflows like compliance reviews; Terra aggregates data from multiple tools for reports; Luna performs high-volume, rule-based tasks efficiently. Two GPT-5.6 models rank in the top 10 of AutomationBench, though they trail newer Anthropic and Google models in overall performance.
Anthropic’s Claude Opus 5 leads the Operations category with a 50.0% success rate on AutomationBench, while Google’s Gemini 3.7 Flash (High) tops the overall leaderboard at 30.44% and costs $0.61 per task. Zapier recommends prioritizing these models for operations-heavy and cost-sensitive workflows, respectively, while noting that additional models remain available within each provider’s integration.
Zapier’s built-in AI by Zapier tool allows users to add AI steps to any workflow without requiring separate accounts. It supports multiple OpenAI and Google models by default and enables model swapping via a dropdown menu, letting teams update workflows quickly when new models outperform existing ones. This feature simplifies template sharing and model preference changes without rebuilding steps.