JMeetEval: leaderboard
Nishika Japanese meeting-transcript instruction following: share of prompts where the model satisfies every attached instruction over 197 synthetic transcripts spanning 16 categories; prompt-level strict accuracy %.
Metric: Prompt-level strict accuracy (%). Source: github.com. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemma-4-E4B-it | 69 | #566 |
| 2 | Qwen 3 VL 8B (Thinking) | 65.5 | |
| 3 | Gemma 4 E2B (IT) | 64 | #781 |
| 4 | Qwen 3 8B | 57.4 | #674 |
| 5 | Qwen3-Swallow-8B-RL-v0.2 | 55.6 | #597 |
| 6 | NVIDIA-Nemotron-Nano-9B-v2-Japanese | 54.8 | #630 |
| 7 | Qwen 3.5 4B | 52 | #475 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=jmeeteval · How It Works · Data refreshed daily, snapshot 2026-10-20.