Best AI model for summarising documents.
Prices checked
gpt-oss-20b at TrustedRouter costs you 0.89¢ per 1,000 pages, the cheapest of 342 models that can do this job, checked .
Cheapest 10
all 392 models- 1
gpt-oss-20b
Resells
TrustedRouter · $0.01 in · $0.05 out0.89¢per 1,000 pages - 2
Llama 3.2 3B Instruct
Runs it themselves
Inference · $0.02 in · $0.02 out1.3¢per 1,000 pages - 3
Llama 3.1 8B Instruct
Runs it themselves
Helicone · $0.02 in · $0.03 out1.4¢per 1,000 pages - 4
Mistral Nemo
Lower precision (fp8)
Eden AI · $0.02 in · $0.03 out1.6¢per 1,000 pages - 5
Ling 3.0 Flash
Same as maker
OpenRouter · $0.02 in · $0.06 out1.6¢per 1,000 pages - 6
Qwen3.7 Flash
Regional priceEden AI · $0.02 in · $0.08 out1.7¢per 1,000 pages - 7
Granite 4.0 H Micro
Same as maker
Cloudflare Workers AI · $0.02 in · $0.11 out1.7¢per 1,000 pages - 8
Nex N2.5 Mini
Same as makerOpenRouter · $0.03 in · $0.10 out2.1¢per 1,000 pages - 9
DeepSeek V4.1 Flash 1017
Runs it themselvesEngy · $0.03 in · $0.12 out
2.5¢per 1,000 pages - 10
Qwen Flash
Regional price
Alibaba Cloud (China) · $0.02 in · $0.22 out2.6¢per 1,000 pages
Other ways to pick
What matters
- 128k context or more
- A 300-page report fits in one call at 128k tokens. Below that you split the document up and pay to re-read the overlap on every piece.
- Input price decides it
- Ten tokens go in for every one that comes out, so a model with a cheap input price and a dear output price is the right shape for this work.
- Reasoning is optional
- Summarising is not a puzzle. A reasoning model bills its thinking as output tokens, which is money spent on the half of the job that is already cheap.
How we count
Reports, contracts and transcripts go in whole and come back as a paragraph. The bill is almost all input tokens, so the input price is the number that decides this job. We count only models with 128k of context or more, and 342 of the 392 we track qualify. The cheapest of them costs 0.89¢ per 1,000 pages at TrustedRouter. They are ranked by what the work costs, not by the price per million tokens, because a model that is cheap to prompt and dear to answer looks different once you know the shape of the job.
A page is about 600 tokens in and 60 out. A cost per 1,000 pages is that multiplied out at the row's own input and output rate, so it says what the work costs; the price per million tokens under each name says what the model costs.
A price below the maker's is listed only when the provider says why.
- Resells
- The provider sells another company's endpoint, at a margin of its own or at the maker's price.
- Runs it themselves
- The provider runs the open weights on hardware of its own, so the price is its own rather than a resale.
- Lower precision
- The open weights, served at a lower precision than the maker's, so not quite the same product.
- Same as maker
- Sold by someone else at the price the maker charges.
- Regional price
- A price list for one region, lower than the international one. A region, not a deal.
Frequently asked
- What is the cheapest model for summarising documents?
- gpt-oss-20b at TrustedRouter, 0.89¢ per 1,000 pages at $0.01 in and $0.05 out per million tokens, read on 31 Oct 2026.
- How is the cost per unit worked out?
- A page is about 600 tokens in and 60 out. Multiply that by 1,000 and price it at each row's own input and output rate. Nothing else is counted: no cache discount, no batch rate, no free tier.
- Why do only 342 models qualify?
- The job needs 128k of context or more. Models that fall short are still in the index, they just cannot do this job as described.
- How current are these prices?
- Prices are read every 3 days from 237 providers' public catalogs and this ranking recomputes with them. Last refresh: 31 Oct 2026.