Best AI for summarising documents / head-to-head
DeepSeek V4 Pro vs GPT-5.3-Codexfor summarising documents
GPT-5.3-Codex wins, 96–95
Same 12 summarising documents tasks, marked blind by three rival labs. GPT-5.3-Codex took 3 tasks, DeepSeek V4 Pro took 3, 6 tied. Tested 13 Aug 2026.
Where they differed most
Summarise in exactly 5 bullets, max 15 words each, readable in 20 seconds: 'Our Q3 revenue rose 12% to £1.4m, driven mainly by the new retail channel which contributed £310k. Margins slipped from 41% to 37% because of higher shipping costs and a one-off warehouse move. Headcount stayed flat at 34. The main risk into Q4 is the supplier contract, which expires in November and has not been renegotiated. Cash at bank is £620k.'
- Revenue rose 12% to £1.4m, driven by new retail channel adding £310k. - Margins fell from 41% to 37% due to higher shipping costs and warehouse move. - Headcount remained unchanged at 34 employees. - Key Q4 risk: supplier contract expires in November, not yet renegotiated. - Cash at bank stands at £620k.
- Q3 revenue increased 12% to £1.4m. - New retail channel drove growth, contributing £310k. - Margins fell from 41% to 37% due shipping costs and warehouse move. - Headcount remained unchanged at 34 employees. - Q4 risk: November supplier contract unrenegotiated; cash at bank is £620k.
Task by task
| Task | DeepSeek V4 Pro | GPT-5.3-Codex |
|---|---|---|
| Meeting notes to actions | 10 | 9 |
| Five-bullet summary | 10 | 8 |
| Do not invent | 10 | 10 |
| One-sentence gist | 8 | 10 |
| Preserve the caveat | 10 | 10 |
| Summarise a disagreement | 10 | 9 |
| Extract the decision | 10 | 10 |
| Long to short, no loss | 10 | 10 |
| Summary for a specific reader | 9 | 9 |
| Bullet the risks only | 10 | 10 |
| Headline and standfirst | 8 | 10 |
| Say when there is nothing to summarise | 9 | 10 |
Full receipts: DeepSeek V4 Pro, GPT-5.3-Codex · judges claude-sonnet-5, gemini-3.1-pro-preview, gpt-5.6-terra
Questions people ask
Which is better for summarising documents: DeepSeek V4 Pro or GPT-5.3-Codex?
GPT-5.3-Codex — it scored 96/100 against 95/100 on our 12-task summarising documents suite, winning 3 tasks to 3 with 6 tied. Every answer was marked blind by three judges from three rival AI labs.
How was this tested?
Both models answered the identical published summarising documents tasks. Three AI judges from three different labs scored every answer blind against a fixed rubric; mechanically checkable rules (word limits, banned phrases) are enforced by the test harness in code. The raw outputs and judge verdicts are downloadable.
More summarising documents head-to-heads: Claude Opus 4.8 vs GPT-5.3-Codex · Claude Opus 4.8 vs DeepSeek V4 Pro · GPT-5.3-Codex vs GPT-5.5 · DeepSeek V4 Pro vs GPT-5.5 · GPT-5.3-Codex vs GPT-5.6 Luna · DeepSeek V4 Pro vs GPT-5.6 Luna
Full ranking: Best AI for summarising documents · model pages: DeepSeek V4 Pro, GPT-5.3-Codex