Which AI models can you automate on Zapier? (Opus 4.8, Gemini 3.5 Flash, and more)

Overview

Zapier now lets business teams plug a wide menu of AI models into their automated workflows, spanning OpenAI, Anthropic, Google, and specialty providers like Mistral and DeepSeek. To help users pick the right one, Zapier ranks each model with AutomationBench, its own test of how well models complete multi-step business tasks rather than one-off prompts. Anthropic's Claude Fable 5.0 leads the overall leaderboard, with Claude Opus 4.8 and Google's Gemini 3.5 Flash close behind.

Key Takeaways

Zapier supports models from OpenAI, Anthropic, and Google, plus specialty providers such as Mistral AI, DeepSeek, Grok, Groq, and OpenRouter, all usable inside the same Zap workflows.
AutomationBench is Zapier's benchmark that scores how well a model handles real end-to-end business tasks, not static prompts, using 47 real tools across six business functions.
Anthropic's Claude Fable 5.0 tops the overall leaderboard at 17.4 percent task completion, ahead of Claude Opus 4.8 and Gemini 3.5 Flash.
Different models win different jobs: GPT-5.5 leads sales and marketing, Fable 5.0 leads operations and finance, and Opus 4.8 leads support.
The AI by Zapier tool lets users swap models from a dropdown without rebuilding the workflow step, making it easy to test which model performs best.

Stats & Key Facts

#Fable 5.0 (Max) leads AutomationBench with a 17.4 percent task completion score.
#Claude Opus 4.8 (XHigh) scores 15.5 percent, and Gemini 3.5 Flash (Medium) scores 14.5 percent on the overall leaderboard.
#AutomationBench tests models against 47 real tools spread across six business functions.
#The benchmark is built on patterns drawn from 2 billion monthly tasks across 3.7 million companies.
#Anthropic and Google models reach context windows of 1 million to 2 million tokens, enough to process very long documents in one pass.
#Fable 5.0 (Max) tops operations at 27.0 percent, the highest single-domain score reported.

The Three Major AI Providers You Can Automate on Zapier

Most workflows draw from three large model families, each with a range of options from heavy reasoning to fast and cheap.

›OpenAI (ChatGPT): GPT-5.5 Pro for the deepest reasoning, GPT-5.5 for complex professional work, GPT-5.4 nano for high-volume repeat tasks, and GPT-4o for handling text, images, audio, and video.
›Anthropic (Claude): Fable 5.0 for the most demanding agentic work, Opus 4.8 for complex reasoning and coding, Sonnet 4.6 for a balance of price and performance, and Haiku 4.5 for low-cost, fast tasks.
›Google (Gemini): Gemini 3.5 Flash for multi-step workflows at scale, Gemini 3.1 Pro for complex reasoning, Gemini 3 Pro for balanced work, and Gemini 2.0 Flash Lite for basic sorting and extraction.

How AutomationBench Scores Models on Real Business Work

Zapier built its own test because standard AI benchmarks measure single prompts, not the chains of steps a real automation runs.

AutomationBench checks whether a model can finish a job from start to end inside realistic business systems. It tests agents against 47 real tools across six functions: sales, marketing, operations, support, finance, and HR. The scenarios are grounded in patterns from 2 billion monthly tasks across 3.7 million companies, so they reflect work that real teams actually do.

The test raises difficulty on purpose by adding irrelevant data, unclear instructions, and strict policy rules a model has to follow. Scores look low at first glance because the bar is hard. A 17.4 percent task completion rate is the current top result, which shows how much room these systems still have to improve on full end-to-end work.

The Overall Leaderboard: Fable 5.0, Opus 4.8, and Gemini 3.5 Flash

Anthropic models hold the top spots, with Google close behind.

›Fable 5.0 (Max): 17.4 percent, the overall leader.
›Fable 5.0 (XHigh): 16.0 percent.
›Claude Opus 4.8 (XHigh): 15.5 percent.
›Claude Opus 4.8 (Max): 15.4 percent.
›Gemini 3.5 Flash (Medium): 14.5 percent.

Which Model Wins Which Department

No single model wins everything, so Zapier breaks results down by business function.

›Sales and marketing: GPT-5.5 from OpenAI leads both.
›Operations: Fable 5.0 (Max) tops the category at 27.0 percent, the highest single-domain score.
›Finance: Fable 5.0 (Max) leads.
›Support: Claude Opus 4.8 (XHigh) takes the top spot.
›The takeaway is to match the model to the job rather than picking one model for every workflow.

Swapping Models Without Rebuilding Your Workflow

Zapier offers a few ways to connect models so you are not locked into one provider.

You can use direct provider integrations for OpenAI, Anthropic, and Google, or the built-in AI by Zapier tool. The AI by Zapier tool lets you change the model from a dropdown without reconfiguring the step, so testing a different model takes seconds instead of a rebuild.

Beyond the big three, Zapier connects to specialty providers including DeepSeek, Grok, Mistral AI, OpenRouter, Groq, and AssemblyAI, plus Google Vertex AI for enterprise setups. That range lets a business pick a model on cost, speed, or accuracy without leaving the platform.

What This Means for a Non-Technical Business Owner

The practical message is that model choice is now a business decision, not only a technical one.

For a small team, the value here is a shortlist backed by testing instead of marketing claims. If you run support, Zapier's data points to Opus 4.8; if you run operations or finance, it points to Fable 5.0; for sales and marketing, GPT-5.5. You do not have to read every model release to decide.

Because models slot into the same Zap and swap from a dropdown, you can run a cheaper, faster model for simple sorting and reserve a top-tier model for the hard, multi-step tasks. That keeps cost down while still getting strong results where it counts.

Frequently Asked Questions

What is AutomationBench?

It is Zapier's own benchmark that measures how well an AI model completes real, multi-step business tasks rather than single prompts. It tests models against 47 real tools across sales, marketing, operations, support, finance, and HR.

Which AI model ranks highest on Zapier right now?

Anthropic's Claude Fable 5.0 (Max) leads the overall AutomationBench leaderboard at 17.4 percent task completion, with Claude Opus 4.8 and Google's Gemini 3.5 Flash close behind.

Do I have to pick one model for all my automations?

No. Different models lead different jobs, so you can match the model to the task. The AI by Zapier tool also lets you swap models from a dropdown without rebuilding the workflow step.

Which providers besides OpenAI, Anthropic, and Google work on Zapier?

Zapier also connects to specialty providers including Mistral AI, DeepSeek, Grok, Groq, OpenRouter, and AssemblyAI, plus Google Vertex AI for enterprise use.

Why are the benchmark scores so low?

AutomationBench raises difficulty on purpose with irrelevant data, unclear instructions, and strict policy rules. The hard bar is why even the top model finishes only about 17 percent of full end-to-end tasks, which shows how much room these systems still have to grow.

Zapier turns model selection into a guided choice by testing every option on real workflow tasks and letting you swap between them inside the same automation. For business teams, that means picking the model that fits each job instead of betting on a single name.