Claude Model Tier Comparison
The four Claude tiers, adaptive thinking and effort, the tool-use rules that differ from other providers, prompt caching, and where you can run it.
Claude is Anthropic's model family trained with Constitutional AI (CAI) for safety. The late-2026 lineup has four tiers: Claude Fable 5.1 ($10/$50 per million tokens) at the top, Claude Opus 5.5 ($4/$20) as the default for demanding work, Claude Sonnet 5.5 ($2/$10) as the production workhorse, and Claude Haiku 5.5 ($0.10/$0.50) as the economy tier. All have a 1M-token context and 128K output. Claude excels at long document analysis, coding, agentic tool use, and nuanced instruction following.
The pipeline
Each stage in order — click any step to read what it does.
Interactive Pipeline Flow
Architecture diagrams
Claude models by use case and cost tier.
Key design decisions
The trade-offs worth knowing before you build this.
Claude models are exceptional at following complex multi-constraint instructions — maintaining format, tone, persona, and content rules simultaneously. For tasks requiring strict output adherence, it is a strong default — verify on your own prompts.
On current Claude models thinking is adaptive and you tune it with effort. Run agentic coding at high or xhigh, routine chat and classification at low. Effort is worth re-tuning on every model upgrade: defaults differ by model (Opus 5.5 defaults to medium, Sonnet 5.5 to high).
If you have a constant large document (code file, knowledge base, long system prompt) that you send with every request, Anthropic's prompt caching drops that context cost by 90%. Cache TTL is 5 minutes by default, or 1 hour on request.
Continue reading
All LLM Model Landscape articles →GPT: OpenAI's Model Family in Practice
The GPT-6 tiers, what reasoning effort replaced, when to use the Responses API over Chat Completions, how the bill is calculated, and what is different on Azure.
Llama: Open-Weight Models for Production
The Llama lineup and licence, serving with Ollama and vLLM, GPU memory sizing, and an honest look at when a hosted API is still the better choice.
Evals for Choosing a Model
Why public benchmarks do not transfer to your workload, how to build a model-selection eval, and why cost per successful task beats price per token.
Discussion
Sign in to share your feedback and join the discussion.