CAST (Discourse Stress) - Text-Only Hit: leaderboard

CAST (Turetzky et al., 2026) pairs one sentence with two contexts that call for stressing different words; in its text-only variant a language model names the word to stress, zero-shot. This board is the share of the 226 items where the named word is the intended one, in percent; higher is better.

Metric: Hit rate (%): items where the predicted stressed word is the intended one; 113 contrastive context pairs (226 items) in which one sentence must stress a different word under each of its two contexts; text-only models name the stressed word zero-shot, non-thinking mode; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Haiku 4.588.1#583
2Qwen 3 4BNon-reasoning57.5#2348
3Qwen 3 1.7BNon-reasoning50#5860
4Qwen 3 0.6BNon-reasoning25.2#8350

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=cast-discourse-stress-text-only-hit · How It Works · Data refreshed daily, snapshot 2026-11-21.