Best LLM for Document Summarisation (2026)

Bottom line up front: For document summarisation, Claude Sonnet 4.6 produces the most faithful, well-structured summaries with the lowest hallucination rate. Gemini 2.5 Pro is the better choice for very long documents that exceed 200K tokens. GPT-4o is a strong alternative for teams that need structured JSON output from their summaries.

What makes a good summarisation LLM

Summarisation seems simple but it exposes model weaknesses quickly:

• Faithfulness — the model must summarise what the document says, not what it thinks the document probably says. Hallucination in summaries is often subtle and difficult to catch
• Context window — you need the model to read the full document in a single pass. Chunking and stitching summaries degrades quality and introduces inconsistencies
• Instruction following — you need to specify format, length, tone, and focus area. Models that drift from these instructions produce summaries that require editing
• Compression ratio — the ability to distil a 50-page document into 3 coherent paragraphs without losing critical information. Not all models do this equally well

Top recommendations

1. Claude Sonnet 4.6 — Best overall for summarisation

Claude Sonnet 4.6 is the best summarisation model available for most document lengths. Its training produces unusually clean compression — it identifies the most important content reliably and presents it in well-formed prose that requires minimal editing. Critically, it has one of the lowest hallucination rates of any frontier model, which is non-negotiable for summaries that will be acted upon.

The 200K context window handles most real-world documents: a 150,000-token limit covers approximately 110,000 words or about 400 pages.

2. Gemini 2.5 Pro — Best for very long documents

Gemini 2.5 Pro's 1M token context window is the primary reason to choose it over Claude for summarisation. If you are processing documents that exceed 200K tokens — full books, complete legal discovery files, large database exports — Gemini 2.5 Pro is the only frontier model that can handle the entire document in a single pass.

Its summarisation quality is slightly below Claude Sonnet 4.6, particularly on documents that require subtle inference or nuanced synthesis. But for documents where the primary challenge is length rather than complexity, it is the correct choice.

It is also significantly cheaper than Claude Sonnet 4.6 on input tokens ($1.25 vs $3.00/M), which matters substantially for very large documents.

3. GPT-4o — Best for structured output summarisation

GPT-4o is the best choice when your summarisation pipeline needs structured output — extracting specific fields, producing JSON with defined keys, or populating a database schema from document content. Its structured output capability is the most reliable of the three models for this pattern.

Its 128K context window is a real limitation for long documents. A 128K context handles approximately 95,000 words — sufficient for most business documents but inadequate for lengthy legal or technical documents.

4. Gemini 2.0 Flash — Best budget option

Gemini 2.0 Flash produces surprisingly good summaries for its price point. For internal summarisation pipelines where summaries are used as inputs to downstream processes (rather than presented directly to users), its quality is often sufficient.

At $0.10/M input, summarising a 50,000-token document costs $0.005 — essentially free at moderate volumes. For high-volume batch summarisation jobs, it is the clear cost winner.

Side-by-side comparison

Cost per document — real-world estimates

FAQ

Last verified: April 2026 · Back to LLM Selector