L2-Bench
Models design lessons, give learners feedback, and participate in language practice. A model judge grades each response against teaching criteria.
- Made by
- James Edgell and colleagues
- Released
- July 2026
- Kind
- Other professions
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Education, Languages and literature
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5.4 #28 on the capability index
- Last model added
- April 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 4.7 leads with 85.5 among models released by April 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 Claude Opus 4.7 API: default 85.5
- 2 GPT-5.4 API: default 84.0
- 3
Gemini 3.1 Pro API: default 83.3
- 4
Gemini 3 Flash API: default 80.6
- 5
DeepSeek-V3.2 API: default 80.1
- 6 Kimi K2.5 API: default 79.0
- 7 Claude Haiku 4.5 API: default 78.6
- 8
Qwen3-32B API: default 65.6
- 9
Magistral Small 1.2 API: default 50.7
Source: Official L2-Bench leaderboard
| # | Model | Score | Setting | Released | Run on | Link | Source |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | 85.5 | API: default | Apr 2026 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 2 | GPT-5.4 | 84.0 | API: default | Mar 2026 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 3 | | 83.3 | API: default | Feb 2026 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 4 | | 80.6 | API: default | Dec 2025 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 5 | | 80.1 | API: default | Dec 2025 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 6 | Kimi K2.5 | 79.0 | API: default | Jan 2026 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 7 | Claude Haiku 4.5 | 78.6 | API: default | Oct 2025 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 8 | | 65.6 | API: default | Apr 2025 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
| 9 | | 50.7 | API: default | Sep 2025 | May 2026 | benchmarks.elt.edu.oup.com | Official leaderboard |
Rankings by part
We read 12 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Course Planning, Lesson Planning and Activity Planning. Claude Opus 4.7 leads 7 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Course Planning
9 models we track have a score here.
- 1 GPT-5.4 ↑1 85.0
- 2 Claude Opus 4.7 ↓1 84.8
- 3
Gemini 3.1 Pro 79.0
- 4 Kimi K2.5 ↑2 78.9
- 5
DeepSeek-V3.2 77.6
- 6 Claude Haiku 4.5 ↑1 76.2
- 7
Gemini 3 Flash ↓3 73.2
- 8
Qwen3-32B 55.8
- 9
Magistral Small 1.2 41.4
Lesson Planning
9 models we track have a score here.
- 1 Claude Opus 4.7 89.7
- 2 GPT-5.4 88.7
- 3
Gemini 3.1 Pro 87.9
- 4 Kimi K2.5 ↑2 87.5
- 5 Claude Haiku 4.5 ↑2 87.2
- 6
DeepSeek-V3.2 ↓1 86.9
- 7
Gemini 3 Flash ↓3 84.5
- 8
Qwen3-32B 73.6
- 9
Magistral Small 1.2 52.3
Activity Planning
9 models we track have a score here.
- 1 Claude Opus 4.7 88.5
- 2
Gemini 3.1 Pro ↑1 86.2
- 3 GPT-5.4 ↓1 85.7
- 4 Claude Haiku 4.5 ↑3 83.7
- 5
DeepSeek-V3.2 83.7
- 6
Gemini 3 Flash ↓2 82.0
- 7 Kimi K2.5 ↓1 79.9
- 8
Qwen3-32B 67.1
- 9
Magistral Small 1.2 47.2
Activity Management
9 models we track have a score here.
- 1 GPT-5.4 ↑1 84.9
- 2 Claude Opus 4.7 ↓1 84.2
- 3
Gemini 3.1 Pro 83.9
- 4
Gemini 3 Flash 82.4
- 5
DeepSeek-V3.2 79.4
- 6 Claude Haiku 4.5 ↑1 75.4
- 7 Kimi K2.5 ↓1 72.5
- 8
Qwen3-32B 61.5
- 9
Magistral Small 1.2 47.8
Language Presentation
9 models we track have a score here.
- 1
Gemini 3.1 Pro ↑2 85.4
- 2 GPT-5.4 84.6
- 3 Claude Opus 4.7 ↓2 84.6
- 4
DeepSeek-V3.2 ↑1 82.9
- 5
Gemini 3 Flash ↓1 81.4
- 6 Claude Haiku 4.5 ↑1 81.0
- 7 Kimi K2.5 ↓1 78.6
- 8
Qwen3-32B 69.6
- 9
Magistral Small 1.2 47.7
Exchange Partner
9 models we track have a score here.
- 1 Claude Opus 4.7 83.8
- 2
Gemini 3.1 Pro ↑1 83.5
- 3
Gemini 3 Flash ↑1 81.7
- 4 Claude Haiku 4.5 ↑3 80.6
- 5 Kimi K2.5 ↑1 79.8
- 6 GPT-5.4 ↓4 78.2
- 7
DeepSeek-V3.2 ↓2 76.9
- 8
Qwen3-32B 70.2
- 9
Magistral Small 1.2 61.4
Performance Evaluation
9 models we track have a score here.
- 1 GPT-5.4 ↑1 80.5
- 2 Claude Opus 4.7 ↓1 80.4
- 3
Gemini 3 Flash ↑1 78.1
- 4
Gemini 3.1 Pro ↓1 77.3
- 5
DeepSeek-V3.2 75.3
- 6 Kimi K2.5 74.6
- 7 Claude Haiku 4.5 73.0
- 8
Qwen3-32B 62.5
- 9
Magistral Small 1.2 46.6
Giving Feedback
9 models we track have a score here.
- 1 Claude Opus 4.7 83.1
- 2
Gemini 3.1 Pro ↑1 81.6
- 3 GPT-5.4 ↓1 78.8
- 4
Gemini 3 Flash 78.3
- 5 Kimi K2.5 ↑1 76.6
- 6
DeepSeek-V3.2 ↓1 75.0
- 7 Claude Haiku 4.5 71.2
- 8
Qwen3-32B 65.0
- 9
Magistral Small 1.2 48.3
Progress Tracking
9 models we track have a score here.
- 1 Claude Opus 4.7 84.7
- 2 GPT-5.4 83.5
- 3
Gemini 3.1 Pro 78.8
- 4
Gemini 3 Flash 78.1
- 5 Kimi K2.5 ↑1 78.1
- 6
DeepSeek-V3.2 ↓1 77.6
- 7 Claude Haiku 4.5 77.0
- 8
Qwen3-32B 66.2
- 9
Magistral Small 1.2 53.6
Emotional Intelligence
9 models we track have a score here.
- 1 Claude Opus 4.7 87.2
- 2 GPT-5.4 84.6
- 3
Gemini 3.1 Pro 84.5
- 4
Gemini 3 Flash 83.9
- 5
DeepSeek-V3.2 82.8
- 6 Claude Haiku 4.5 ↑1 81.5
- 7 Kimi K2.5 ↓1 81.3
- 8
Qwen3-32B 74.8
- 9
Magistral Small 1.2 65.6
Assessment Creation
9 models we track have a score here.
- 1 GPT-5.4 ↑1 82.8
- 2 Claude Opus 4.7 ↓1 82.2
- 3
Gemini 3.1 Pro 80.6
- 4
Gemini 3 Flash 74.5
- 5
DeepSeek-V3.2 73.7
- 6 Kimi K2.5 72.7
- 7 Claude Haiku 4.5 67.7
- 8
Qwen3-32B 43.8
- 9
Magistral Small 1.2 30.4
Teacher Development
9 models we track have a score here.
- 1 Claude Opus 4.7 92.5
- 2
Gemini 3.1 Pro ↑1 91.4
- 3 GPT-5.4 ↓1 91.1
- 4
DeepSeek-V3.2 ↑1 89.8
- 5
Gemini 3 Flash ↓1 89.5
- 6 Claude Haiku 4.5 ↑1 89.1
- 7 Kimi K2.5 ↓1 87.6
- 8
Qwen3-32B 77.0
- 9
Magistral Small 1.2 65.9
The best score, 85.5, costs $0.05 a task
Runs on the cost frontier score higher than any run that's cheaper. Gemini 3.1 Pro at API: default comes within 2.2 of the best for $0.02 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official L2-Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 4.7 | API: default | 85.5 | $0.05 | ||
| GPT-5.4 | API: default | 84.0 | $0.03 | ||
| API: default | 83.3 | $0.02 | |||
| API: default | 80.6 | $0.003 | |||
| API: default | 80.1 | $0.001 | |||
| Kimi K2.5 | API: default | 79.0 | $0.002 | ||
| Claude Haiku 4.5 | API: default | 78.6 | $0.005 | ||
| API: default | 65.6 | $0.0004 | |||
| API: default | 50.7 | $0.0006 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official L2-Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official L2-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Claude Opus 4.7. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (9 entries).
- Pass Models scored 9 models we track.
- Warning Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on benchmarks.elt.edu.oup.com: 9 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.