- type
- summary
- created
- Mon Apr 06 2026 02:00:00 GMT+0200 (Central European Summer Time)
- updated
- Mon Apr 06 2026 02:00:00 GMT+0200 (Central European Summer Time)
- sources
- raw/plans/openrouter-llm-research
- tags
- research llm ai cost-analysis
Summary: OpenRouter LLM Research
abstract
Ranked comparison of 312+ models on OpenRouter API — top pick: Gemini 2.0 Flash ($0.10/$0.40 per M tokens). Cost benchmarks for invoice extraction tasks.Key Findings
- Top pick:
google/gemini-2.0-flash-001— best price/performance for structured extraction - Cost: $0.10 input, $0.40 output per M tokens (~$0.0004 per invoice)
- Alternatives: Qwen3-32B (free tier available), DeepSeek V3.1 (high quality)
- Claude Sonnet used for higher-level tasks (renewal letters, chat) — more expensive but better reasoning
Source
Related
- wiki/entities/openrouter — the API gateway
- wiki/concepts/invoice-processing — primary use case
- wiki/concepts/tech-stack — AI infrastructure