DexHoldem Agentic Perception - Opponent Chip Inventory: leaderboard
Metric: Field accuracy (%; exact per-denomination count of the opponent's chips, 16 problems; 36 tabletop states from real games, mean of three runs; the perceiver parses the image with its harness at a medium thinking budget). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 (Claude Code, Medium) | 43.8 |
| 2 | GPT-5.5 (Codex, Medium) | 35.4 |
| 3 | Claude Sonnet 4.6 (Claude Code, Medium) | 22.9 |
| 4 | Gemini 3 Flash (Gemini CLI, Medium) | 22.9 |
| 5 | GPT-5.4 (Codex, Medium) | 18.8 |
| 6 | GPT-5.4 mini (Codex, Medium) | 18.8 |
| 7 | Claude Haiku 4.5 (Claude Code, Medium) | 18.8 |
| 8 | Gemini 3.1 Flash-Lite (Gemini CLI, Medium) | 14.6 |
Interactive version: theaggregate.ai/benchmark?slug=dexholdem-agentic-perception-opponent-chip-inventory · How It Works · Data refreshed daily, snapshot 2026-09-26.