UNICBench (Text): leaderboard
Metric: Exact-count hit rate (%): share of parsed numeric answers equal to the annotated count over UNICBench's 5,888 text counting questions (872 documents: code, JSON, HTML, MusicXML, LaTeX, CSV, legal, news and literary text up to millions of characters), temperature 0 (1.0 where a model fixes it), 4,096-token cap, GPT-5 and GPT-5-mini at minimal and o3 at low reasoning effort; unparsable responses are left out of the rate; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 63.3 | #145 |
| 2 | O3 (Low) | 61.1 | #121 (O3) |
| 3 | DeepSeek R1 0528 | 59.7 | #217 |
| 4 | O4 Mini | 59.6 | #172 |
| 5 | DeepSeek V3.1 (Thinking) | 47.3 | #260 (DeepSeek V3.1) |
| 6 | DeepSeek V3.1 (Non-reasoning) | 40.1 | #260 (DeepSeek V3.1) |
| 7 | GPT-4o | 39.8 | #333 |
| 8 | Claude Sonnet 4 (20250514) | 37.7 | #211 |
| 9 | Grok 4 Fast (Non-reasoning) | 37.6 | #242 (Grok 4 Fast) |
| 10 | Gemini 2.5 Flash (Non-reasoning) | 37.3 | #237 (Gemini 2.5 Flash) |
| 11 | Qwen 3 235B A22B 2507 Instruct | 36.2 | #291 |
| 12 | GPT-5 (Minimal) | 36.2 | #91 (GPT-5) |
| 13 | GPT-4o Mini | 35.5 | #588 |
| 14 | Seed-OSS-36B-Instruct | 35.3 | #348 |
| 15 | GPT-5 Mini (Minimal) | 35.3 | #176 (GPT-5 Mini) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=unicbench-text · How It Works · Data refreshed daily, snapshot 2026-10-11.