L2-Bench: leaderboard
L2-Bench evaluates how well models perform second-language education work: a thousand expert-reviewed tasks spanning twelve teaching and assessment competencies, each parameterized by learner, context and resource factors. Each response is graded criterion by criterion against a weighted pass or fail rubric with an expert reference answer, and the percentage score is the weighted share of criteria passed, with negative criteria acting as penalties.
Metric: Rubric score (%; weighted share of task criteria passed, negative criteria as penalties and task scores clipped at 0, mean over 1,000 second-language education tasks and three runs each; per-criterion pass/fail verdicts from a reference-guided Claude Sonnet 4.6 judge; provider-default inference settings). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 85.5 | #79 |
| 2 | GPT-5.4 | 84.1 | #123 |
| 3 | Gemini 3.1 Pro (Preview) | 83.4 | #94 |
| 4 | Gemini 3 Flash | 80.7 | #164 |
| 5 | DeepSeek V3.2 | 80.2 | #406 |
| 6 | Kimi K2.5 | 79.1 | #256 |
| 7 | Claude Haiku 4.5 | 78.8 | #568 |
| 8 | Qwen 3 32B | 65.8 | #960 |
| 9 | Magistral Small | 50.7 | #2295 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=l2-bench · How It Works · Data refreshed daily, snapshot 2026-11-13.