InstrEditBench (Multi Turn; Zero Shot) - Code BLEU: leaderboard
BLEU (0-1) against reference edited text; InstrEditBench Code, multi turn, zero shot; 90/10 train-test split. Multi-turn applies requests cumulatively on 100 randomly sampled multi-turn items per domain. FineEdit trained derivatives remain separate from their Llama/Qwen bases; higher is better.
Metric: BLEU (0-1) against reference edited text; InstrEditBench Code, multi turn, zero shot; 90/10 train-test split. Multi-turn applies requests cumulatively on 100 randomly sampled multi-turn items per domain. FineEdit trained derivatives remain separate from their Llama/Qwen bases; higher is better.. Source: aclanthology.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 3B Instruct | 0.4 | #1149 |
| 2 | Gemini 1.5 Flash | 0.39 | #623 |
| 3 | Llama 3.2 3B | 0.3 | #1380 |
| 4 | Llama 3.2 1B | 0.17 | #1545 |
| 5 | Gemini 2.0 Flash | 0.14 | #332 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=instreditbench-multi-turn-zero-shot-code-bleu · How It Works · Data refreshed daily, snapshot 2026-10-16.