Best LLM API for summarization & data extraction
Summarization and extraction are typically input-heavy (a long document in) and output-light (a short summary or a structured field out), so the input rate drives cost. Budget-tier models, cheapest input first.
Summarization and structured-extraction tasks share a pattern: a long input and a comparatively short, structured output. That makes them input-price-sensitive in the same way high-volume classification tasks are. This list filters to budget-tier models sorted by input price ascending.
| Model | Provider | Input /1M | Output /1M | Context |
|---|---|---|---|---|
| Mistral | $0.15 | $0.60 | Not published | |
| Meta (via Together AI) | $0.18 | $0.59 | 1M tokens | |
| OpenAI | $0.20 | $1.20 | ~1.05M tokens | |
| $0.25 | $1.50 | Not published | ||
| Meta (via Together AI) | $0.27 | $0.85 | ~1.05M tokens | |
| DeepSeek | $0.30 | $1.20 | 1M tokens | |
| Amazon | $0.30 | $2.50 | 1M tokens | |
| Alibaba | $0.40 | $1.60 | 1M tokens | |
| Mistral | $0.50 | $1.50 | Not published | |
| Anthropic | $1.00 | $5.00 | 200K tokens | |
| xAI | $1.00 | $2.00 | 256K tokens |
We don't benchmark task-specific quality — this shortlist is built from verified price and published context window only. Use it to narrow candidates by cost, then evaluate output quality yourself.
Frequently asked questions
The filter and sort are the same because the cost pattern overlaps — both are input-heavy, output-light workloads. This page exists separately so summarization and extraction are easy to find on their own.
Yes — if your documents regularly exceed a model's context window, you'll need to chunk them, which adds complexity and can hurt summary quality across chunk boundaries. Check the context column, not just price.
Yes — we track verified price and context window only, not extraction accuracy. For fields that matter, validate a sample of outputs against ground truth before relying on any model at scale.