NCP-Bench - Satisfied Commitments: leaderboard

Metric: Satisfied commitments (%; share of the narrative commitments in the Satisfied state at termination, averaged over the environments; 100 movie-synopsis narrative environments, each with an initial fact ledger, reference trajectory and commitment set; the evaluated model narrates against an adversarial Gemini-2.5-Flash player, and Gemini-2.5-Flash auditors check every narrator response and end the episode at the first confirmed fact, commitment or player-input conflict, at 100 turns, or when every achievement commitment is met; temperature 0.6, top-p 0.95). Source: arxiv.org. Saturation forecast: Around 2033. 6 models tracked.

Top models

#ModelScore
1DeepSeek V3.213.42
2GPT-5.211.22
3GPT-4o Mini10.9
4Qwen 3 235B A22B10.6
5Grok 4.1 Fast10.37
6Kimi K2.55.18

Interactive version: theaggregate.ai/benchmark?slug=ncp-bench-satisfied-commitments · How It Works · Data refreshed daily, snapshot 2026-09-29.