Anthropic's Claude family has expanded to include four publicly available models, Haiku, Sonnet, Opus, and Fable, plus a fifth model called Mythos that remains unavailable to the general public. The lineup spans a wide range of capabilities and price points, but recent testing reveals that the relationship between model size and real-world performance is far more nuanced than many assume.

The version numbers across Claude's family are inconsistent. As of late September 2026, Opus sits at version 5.5, Fable reached 5.1 on September 1, 2026, Sonnet launched version 5 on June 30, 2026, and Haiku remains on 4.5 since its October 2025 debut, though a new version is expected soon.

How Do Claude's Four Main Models Compare in Real-World Tasks?

• Claude Haiku 4.5: The fastest and cheapest option at $1 per million input tokens and $5 per million output tokens. Best suited for high-volume, repeatable tasks like message classification and ticket routing. Haiku correctly categorized all 25 customer support messages in testing but struggled with reasoning tasks and sometimes failed silently without alerting users to errors.
• Claude Sonnet 5: Priced at $2 per million input tokens and $10 per million output tokens, Sonnet serves as the everyday default for chat, drafting, and analysis work. It's the most advanced model available on Anthropic's free plan and showed strong performance on document analysis and rewriting tasks, though it missed the word count target by 17 words in one test.
• Claude Opus 5.5: Available only to paid subscribers at $4 per million input tokens and $20 per million output tokens, Opus offers the best balance of price and performance for coding and development work. It excels at reasoning tasks and correctly identified unsolvable problems instantly, though it sometimes applies business logic that differs from documented metrics.
• Claude Fable 5.1: The frontier model for advanced coding and long-running agentic work, priced at $10 per million input tokens and $50 per million output tokens. Fable includes a fallback mechanism to Opus on flagged requests and represents Anthropic's most capable offering, though it requires Pro or Max plan access.

All four Claude models demonstrate remarkable capability. Even Haiku, the smallest, outperforms the best frontier models from two years ago, making direct performance comparisons challenging.

What Does Better Data Access Do for Smaller Models?

A critical finding emerged from testing how Claude models interact with data infrastructure. When researchers benchmarked Claude Haiku and Opus against the same retail data questions, Haiku initially produced incorrect answers on join operations, while Opus applied business logic that didn't match the defined metrics.

The breakthrough came when the researchers tested five different interfaces to the same data project. Requiring Haiku to read documentation before querying reduced its wrong answers from four to zero across 150 runs. The Semantic Layer, a governed interface that defines metrics once and enforces consistency, brought both Haiku and Opus to zero wrong answers while delivering the lowest cost per trusted answer for both models.

"The interface mattered significantly. Requiring Haiku to read the dbt documentation before querying took its wrong answers on the join questions from 4 to 0. The Semantic Layer was the only interface that brought both models to 0 wrong answers, and it did so at the lowest cost per trusted answer for both," noted Kelly Kohlleffel, Senior Global Director of Partner Sales Engineering at Fivetran.

Kelly Kohlleffel, Senior Global Director of Partner Sales Engineering at Fivetran

This finding reshapes how organizations should think about deploying AI agents. Better-governed access to data can matter as much as the model itself, meaning that companies might achieve superior results by pairing a smaller, cheaper model with a well-structured data layer rather than upgrading to a larger model without improving data governance.

Which Claude Model Should You Actually Use?

The choice depends heavily on the task and budget constraints. For simple, repeatable work like categorizing messages or routing support tickets, Haiku delivers strong results at minimal cost. However, Haiku is best deployed through an API for automated tasks rather than for daily interactive use, since it can fail silently without alerting users to problems.

Sonnet strikes a practical middle ground. It's capable enough for most linguistic reasoning tasks, produces working code, and costs significantly less than Opus. For users on Anthropic's free plan or those looking to conserve token usage, Sonnet represents the most advanced option available.

Opus remains the default for most coding and development work. It handles reasoning tasks with clarity, correctly identifies unsolvable problems, and provides the best balance of capability and cost for professional use. Fable, meanwhile, targets frontier applications requiring the absolute highest performance, though its 5x price premium over Opus limits it to specialized use cases.

The broader lesson from recent benchmarking is that model selection is no longer a simple matter of choosing the largest or most expensive option. Context matters. Data governance matters. Task specificity matters. Organizations that understand these nuances can deploy Claude models far more efficiently than those that default to the largest available model for every problem.