Arena Releases Statistics on Model Assertions
arena · x · 2026-07-16
Arena explained how they annotate model assertions and how these assertions are distributed across battles.
- They have annotated assertions from over 2 million real LLM conversations: over 1.3 million in Text Arena and over 700,000 in Search Arena.
- This data covers approximately 130,000 Text Arena battles and 40,000 Search Arena battles.
- In Text Arena, the proportion of replies with at least one assertion from either side is 76%; in Search Arena, it is 88%.
- On average, there are about 5 assertions per reply in Text Arena, and nearly 10 in Search Arena.
- The post also mentions that the marginal true assertion rate is 87% in Text Arena and 89% in Search Arena. This relates to Arena's AutoModality classifier, which automatically routes prompts that are more likely to require internet access to Search Arena.
Related event: Arena Adds Factuality to Model Rankings(12 posts)→
More from Models
- Kimi K2.8 Preview rolls out: near-K3 coding performance, 1M context for all tiers — teortaxesTex · 2026-09-11
- Looking for a classifier of software engineering task shapes to pick models per task — StewartalsopIII · 2026-09-11
- DeepSeek V4 Pro API to continue after Sept 2026, billing unchanged — teortaxesTex · 2026-09-11
- DeepSeek V4.1 Flash Hits 98% of GPT-6 Astra's Score at 1.4% of the Cost in Third-Party Benchmark — ayushtweetshere · 2026-09-11
- TheZvi Polls: Has Your Coding Model Choice Changed Since Fable 5.1 and Astra? — TheZvi · 2026-09-11
- antirez Weighs In on Anthropic Banning Minors From Using Claude — antirez · 2026-09-11