Anthropic's Claude 5 family spans several models at different capability and cost points. Here's how Fable 5.1 fits, and a method for picking the right model instead of defaulting to the biggest one.
Anthropic's Claude 5 generation is a family rather than a single model. Each tier trades capability against latency and cost, and the practical skill is picking the cheapest tier that clears your quality bar — not reaching for the largest one by default.
| Model | Positioning | Typical use |
|---|---|---|
| Claude Opus 5 | The frontier tier — deepest reasoning | Hard multi-step problems, long agentic runs, complex code, research and analysis |
| Claude Sonnet 5 | The balanced workhorse | Most production workloads: coding, writing, extraction, chat at scale |
| Claude Fable 5.1 | Part of the Claude 5 family, offered alongside the tiers above | Available as a selectable model where the Claude 5 family is offered |
| Claude Haiku 4.5 | The fast, inexpensive tier | High-volume classification, routing, simple extraction, latency-sensitive paths |
Three questions settle it for most teams:
If a wrong answer gets caught by a human two seconds later, use the cheap fast model. If a wrong answer ships to a customer, gets committed to a repository, or compounds across twenty subsequent steps, pay for the frontier tier. Cost per token is the wrong unit; cost per incident is the right one.
Short, bounded tasks — classify this, extract that, rewrite this paragraph — are where smaller models are closest to frontier models and where the price difference is pure savings. The gap widens sharply as tasks get longer and require holding more state.
A user waiting on a chat response has a very different tolerance than a batch job running overnight. Route interactive paths to fast models and escalate to a larger model only when the fast one signals uncertainty.
The single highest-return thing you can do is assemble 30–50 real examples from your own workload with known-correct answers. Then:
This takes an afternoon and it replaces every model-comparison article on the internet, including this one, with an answer specific to your actual work. It also means the next model release is a config change and a test run instead of a research project.
Model releases are frequent and the ranking changes. Put a thin abstraction between your application and whichever model you call, keep prompts in version control, and keep the eval suite runnable with one command. Teams that do this adopt new models in a day. Teams that don't spend a quarter on it.
For a cross-vendor view, see our comparison of Claude, GPT, Gemini and the leading Chinese models.
Pick the tool that does the job, not the one with the biggest spec sheet. XAIWebsites generates a complete, professional business website — content, layout, mobile version and hosting — from a few questions.
Published 11 September 2026 by the XAIAgent editorial team. Rumor-tracker pages cover unannounced products; we label what is confirmed, what is reported, and what is expectation. See our methodology.