Best LLMs for Document Summarization
Large inputs, short outputs — input rate and context window decide the bill.
Typical request this page prices
- Quality floor: Top 50% of priced models by Intelligence Index (≥ 14.9).
- Only models with a recorded context window of at least 128K tokens.
- Costs use each provider’s current standard rates, with cached tokens at the cached-input rate where one is published. Speed is from Artificial Analysis where measured.
Lowest cost that meets the bar
Models that clear the quality floor, ordered by estimated cost for the request above.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| Gemma 4 26B Google · 262K context | 16.7 | 101tok/s | OpenRouter 7 providers | $0.042 / $0.220 cached $0.021 | $1.23 |
| DeepSeek V4 Flash DeepSeek · 1.0M context | 34.3 | - | OpenRouter 19 providers | $0.013 / $1.28 cached $0.013 | $1.35 |
| Step 3.5 Flash Stepfun · 262K context | 17 | - | Runcrate 3 providers | $0.090 / $0.300 | $2.49 |
| Qwen 3.5 35B Alibaba · 262K context | 19.3 | - | OpenRouter 6 providers | $0.080 / $0.750 cached $0.040 | $2.60 |
| Gemini 3.1 Flash Lite Google · 1.0M context | 15.6 | - | Velokey 5 providers | $0.100 / $0.600 cached $0.010 | $2.98 |
| MiniMax M2.7 MiniMax · 205K context | 22.8 | - | Empirio Labs 9 providers | $0.150 / $0.600 cached $0.030 | $4.23 |
| Gemini 2.5 Flash Google · 1.0M context | 15.5 | - | Runcrate 5 providers | $0.150 / $1.25 | $4.75 |
| Gemini 3 Flash Google · 1.0M context | 26.3 | - | Velokey 4 providers | $0.200 / $1.20 cached $0.020 | $5.96 |
Highest-scoring options
Ordered by Artificial Analysis Intelligence Index, with the cost of the same request.
| Model | Intelligence | Speed | Cheapest host | Input / output per 1M | Per 1,000 requests |
|---|---|---|---|---|---|
| Claude Fable 5 Anthropic · 1.0M context | 49.6 | - | Velokey 6 providers | $4.00 / $20.00 cached $0.400 | $116.00 |
| Qwen 3.8 Max Alibaba · 1.0M context | 45.4 | 36tok/s | Deep Infra 9 providers | $1.65 / $4.95 cached $0.206 | $45.21 |
| Kimi K3 Moonshot · 1.0M context | 43.6 | 40tok/s | OpenRouter 16 providers | $0.730 / $10.00 cached $0.550 | $26.25 |
| Claude Opus 4.8 Anthropic · 1.0M context | 41.8 | - | Velokey 6 providers | $2.00 / $10.00 cached $0.200 | $58.00 |
| Claude Opus 4.7 Anthropic · 1.0M context | 40.7 | - | Velokey 6 providers | $2.00 / $10.00 cached $0.200 | $58.00 |
| Gemini 3.7 Flash Google · 1.0M context | 39.6 | - | Deep Infra 5 providers | $0.750 / $3.75 | $21.75 |
| GPT-5.4 OpenAI · 128K context | 39 | - | Velokey 6 providers | $0.750 / $4.50 cached $0.075 | $22.35 |
| DeepSeek V4 Pro DeepSeek · 1.0M context | 36 | 89tok/s | Velokey 17 providers | $0.435 / $0.870 cached $0.0036 | $11.57 |
Compare every model and provider on the LLM API pricing table.
How to choose an LLM for document summarization
Summarization reads a lot and writes a little. A report, transcript or contract runs to tens of thousands of tokens while the summary is a few hundred, so the input rate is almost the whole cost. Each document is usually different, so prompt caching helps little and the estimate assumes none.
The model must hold the full document in its context window. This shortlist only includes models with a published context window of at least 128K tokens; models whose window is not recorded in our catalog are left out rather than assumed to qualify.
Summarization is rarely interactive. If results can wait, batch APIs from several providers bill the same models at a discount, which the per-model pages show separately from standard rates.
Frequently Asked Questions
Why are some recent models missing from this list?
The list only includes models with a recorded context window of at least 128K tokens. Some newer models do not yet have their context window recorded in our catalog, and they are excluded until it is, rather than being assumed to fit long documents.
Is it cheaper to chunk a document than to send it whole?
Chunking with a map-reduce pass sends roughly the same input tokens plus extra output for intermediate summaries, so it rarely saves money. It can improve quality on very long inputs where models lose track of the middle of the context.
When should I use batch pricing?
For backfills and scheduled jobs where results are not needed within seconds. Batch endpoints typically return within hours and are billed below standard rates; compare them on each model’s page.