what LLM API should I use for extraction

If your goal is structured extraction from text (entities, fields, records, JSON), I’d usually recommend:

Best default choice

OpenAI GPT-4.1 / GPT-4o with structured output / JSON schema support

• Very strong at following extraction instructions
• Good accuracy on messy, real-world text
• Reliable for producing valid JSON
• Easy to use for schema-based extraction

Good alternatives

• Anthropic Claude: often excellent for long documents and nuanced extraction
• Google Gemini: good if you already use Google Cloud / need multimodal extraction
• AWS Bedrock models: good if you want enterprise deployment flexibility

What matters most for extraction

Pick the API that supports:

1. JSON / schema-constrained output
2. Low hallucination behavior
3. High context length if documents are long
4. Tool/function calling if you want validation or downstream processing

Practical recommendation

• For most teams: OpenAI GPT-4.1 or GPT-4o
• For long, complex docs: Claude
• For cost-sensitive large-scale pipelines: test smaller models with a validation layer

Important tip

For extraction, don’t rely on “just prompt it.” Use:

• a strict schema
• retry/validation logic
• post-processing checks

If you want, I can recommend the best model/API for your specific extraction task if you tell me:

• what you’re extracting
• document type
• volume
• latency/cost constraints