AI / Claude Models Basics Interview Questions
What are the key differences between Claude 4 and earlier Claude 3 generation models?
Claude 4 (and the Claude 4/5 generation more broadly) represents significant advances over Claude 3 across capability, context, and new features. Understanding what changed helps teams make informed migration decisions.
| Feature | Claude 3 generation | Claude 4+ generation |
|---|---|---|
| Flagship model | Claude 3 Opus | Claude Opus 4.8, Fable 5 |
| Context window | 200,000 tokens (Opus 3 max) | 1 million tokens (Opus 4.8, Sonnet 5, Fable 5) |
| Thinking | Not available | Extended thinking (Haiku 4.5) and Adaptive thinking (Opus/Sonnet) |
| Tool streaming | Beta feature | GA on Sonnet 4.6+, no beta header needed |
| Computer use | Preview on 3.5 Sonnet | GA on all Claude 4+ models; updated tool versions |
| Effort parameter | Not available | Available on Opus 4.8 and 4.7 |
| Model ID format | claude-3-opus-20240229 | claude-opus-4-8 (no date suffix for newer models) |
| Extended output | Beta | GA on selected models (300k via Batches API) |
| Vision (images per request) | Up to 20 images | Up to 600 images (100 for 200k window models) |
Migration compatibility: Claude 4+ models use the same Messages API as Claude 3. Most Claude 3 code is compatible with Claude 4 models with just a model ID change. Key things to test after migration: tool call formatting, thinking feature support, and any model-specific prompt tuning that assumed Claude 3 response patterns.
What is the biggest context window improvement from Claude 3 to Claude 4+ (comparing flagship models)?
When migrating from a Claude 3 model to a Claude 4+ model, what is typically the minimum code change required?
More Related questions...
