Benchmark Almanac

L2-Bench

Claude Opus 4.7 85.5 leads since July 2026
Not in model cards

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
In and out
Text → Text
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
9
Models scored on its own leaderboard

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:

  1. 1 Claude Opus 4.7 API: default 85.5
  2. 2 GPT-5.4 API: default 84.0
  3. 3 Gemini 3.1 Pro API: default 83.3
  4. 4 Gemini 3 Flash API: default 80.6
  5. 5 DeepSeek-V3.2 API: default 80.1
  6. 6 Kimi K2.5 API: default 79.0
  7. 7 Claude Haiku 4.5 API: default 78.6
  8. 8 Qwen3-32B API: default 65.6
  9. 9 Magistral Small 1.2 API: default 50.7

Source: Official L2-Bench leaderboard

#ModelScoreSettingReleasedRun onLinkSource
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 Gemini 3.1 Pro 83.3 API: default Feb 2026 May 2026 benchmarks.elt.edu.oup.com Official leaderboard
4 Gemini 3 Flash 80.6 API: default Dec 2025 May 2026 benchmarks.elt.edu.oup.com Official leaderboard
5 DeepSeek-V3.2 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 Qwen3-32B 65.6 API: default Apr 2025 May 2026 benchmarks.elt.edu.oup.com Official leaderboard
9 Magistral Small 1.2 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. 1 GPT-5.4 ↑1 85.0
  2. 2 Claude Opus 4.7 ↓1 84.8
  3. 3 Gemini 3.1 Pro 79.0
  4. 4 Kimi K2.5 ↑2 78.9
  5. 5 DeepSeek-V3.2 77.6
  6. 6 Claude Haiku 4.5 ↑1 76.2
  7. 7 Gemini 3 Flash ↓3 73.2
  8. 8 Qwen3-32B 55.8
  9. 9 Magistral Small 1.2 41.4

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.

Qwen DeepSeek Google DeepMind OpenAI Anthropic Cost frontier One model at each effort level
0.00 20.0 40.0 60.0 80.0 100.0 $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale Claude Opus 4.7, API: default Gemini 3 Flash, API: default Qwen3-32B, API: default 0.00 20.0 40.0 60.0 80.0 100.0 $0.0001 $0.001 $0.01 $0.10 Cost per task, log scale Claude Opus 4.7, API: default Qwen3-32B, API: default

Source: Official L2-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.7 API: default 85.5 $0.05
GPT-5.4 API: default 84.0 $0.03
Gemini 3.1 Pro API: default 83.3 $0.02
Gemini 3 Flash API: default 80.6 $0.003
DeepSeek-V3.2 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
Qwen3-32B API: default 65.6 $0.0004
Magistral Small 1.2 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.

Anthropic
Own leaderboard Not tested since Jul 2026

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.

Plot with other benchmarks
On the benchmark’s own leaderboard
50.0 60.0 70.0 80.0 90.0 2025 2026 50.0 60.0 70.0 80.0 90.0 2025 2026

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

All 637 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.

How we check