Claude models: Fable vs. Opus vs. Sonnet vs. Haiku

Anthropic’s Claude family includes four publicly available models: Haiku, Sonnet, Opus, and Fable, each with varying version numbers and release dates. Opus 5.5 was the latest at the time of writing, while Fable 5.1 launched on September 1, 2026, Sonnet 5 on June 30, 2026, and Haiku remains on version 4.5 since October 2025, with an update expected soon. Mythos is another model in the family but is not accessible to the general public. All four models demonstrate strong capabilities, with Haiku outperforming frontier models from two years prior, complicating direct comparisons.

Zapier’s AutomationBench evaluates AI agents on end-to-end workflows using real tools across business functions, simulating tasks like CRM updates and calendar management. Built on patterns from over 2 billion monthly tasks across 3.7 million companies, the benchmark provides a standardized way to measure model performance. The tests reveal nuanced differences in how each model handles complex, multi-step processes, offering insights beyond traditional benchmarking methods.

To assess the models independently, the author used identical prompts across five tasks: categorizing 25 customer support messages, solving an impossible calendar puzzle, rewriting a product announcement, analyzing documents for contradictions, and building a Kanban board in a single HTML file. These tests focused on reasoning, adaptability, and precision, highlighting strengths and weaknesses not captured by standard metrics.

The author conducted the tests using the Claude app but noted that all models are available via Claude Code and the API, with Claude Code recommended for advanced use cases. The models operate with a 200K context window and a 64K maximum output, though Haiku is still on version 4.x with an update pending. Usage varies by plan, with Free offering 1M context and 128K output, Pro including usage credits, and Max providing 50% of weekly limits before falling back to Opus on flagged requests.