Benchmark Almanac

K12-Bench

Gemini 3 Flash 57.1% leads since May 2026
Not in model cards

Models select curriculum concepts, prerequisite relationships, related knowledge, experimental evidence, and textbook locations. The 23,640 multi-select questions come from a textbook knowledge graph. Exact match and instance-level F1 compare selected options with graph-derived answers.

K12-Bench is the evaluation component of K12-KGraph. K12-Train is a separate training corpus. Questions reflect Chinese textbook organization across mathematics, physics, chemistry, and biology.

Released
May 2026
Kind
Language reasoning
Level
School
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
Chinese
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-5.2 #51 on the capability index
Last model added
April 2026
Scores refreshed
November 19th, 2026

In the authors’ words

Large language models are increasingly used in K-12 education, but existing benchmarks mainly test exam question answering rather than understanding how curriculum knowledge is structured and visually presented. We call this capability curriculum cognition. It covers prerequisite chains, concept taxonomies, experiment-concept links, pedagogical sequencing, and visual grounding. We introduce K12-KGraph, a curriculum-aligned knowledge graph extracted from official People's Education Press textbooks in mathematics, physics, chemistry, and biology across primary, middle, and high school. It contains nine node types and fourteen relation types covering curriculum structure and visual grounding. From this graph, we derive K12-Bench, a 23,640-question multi-select benchmark with five task families: Ground, Prereq, Neighbor, Evidence, and Locate. We also build K12-Train, a graph-guided supervised fine-tuning corpus of 7,335 samples, including 2,267 text-only QA pairs and 5,068 multimodal VQA pairs. On K12-Bench, Gemini-3-Flash achieves only 57 percent exact match and Gemma-4-31B-IT reaches 46 percent, with Prereq and Neighbor being the hardest tasks. Our training experiments show that domain-specific supervision can reduce this gap. Under a matched 2,300-sample budget, K12-Train-Text consistently outperforms equally sized subsets of eight mainstream instruction-tuning corpora on GaokaoBench and EduEval. For vision-language models, K12-Train-Full achieves the best overall results on Gaokao-MM, MDK12-medium, and K12Vista among all compared training configurations, despite using fewer samples than the full DataFlow and WizardLM baselines. It also surpasses both text-only and multimodal-only variants, showing that textual and visual supervision are complementary. We release the graph, benchmark, training data, and complete construction pipeline.

Abstract of K12-KGraph: A Curriculum-Aligned Knowledge Graph for Benchmarking and Training Educational LLMs (arXiv:2605.09635, CC0)
8
Entries on its own leaderboard, 6 models we track

Gemini 3 Flash leads with 57.1% among models released by April 2026 Gemini 3 Flash leads with 57.1% among models released by April 2026

Best score per entry from the benchmark's own leaderboard, read on November 19th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

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 Gemini 3 Flash 57.1%
  2. 2 Gemini 2.5 Flash Sep 2025 48.3%
  3. 3 Gemma 4 31B IT 46.4%
  4. 4 GPT-5.2 42.8%
  5. 5 Qwen3-32B 42.6%
  6. 6 glm-4.7-flash 31.7%
  7. 7 GPT-4o (snapshot unspecified) 31.1%
  8. 8 Meta-LLaMA-3-8B-Instruct 7.20%

Source: Official K12-Bench leaderboard

#EntryScoreSettingDateSource
1 Gemini 3 Flash 57.1% Dec 2025 Official leaderboard
2 Gemini 2.5 Flash Sep 2025 48.3% Sep 2025 Official leaderboard
3 Gemma 4 31B IT 46.4% Apr 2026 Official leaderboard
4 GPT-5.2 42.8% Dec 2025 Official leaderboard
5 Qwen3-32B 42.6% Apr 2025 Official leaderboard
6 glm-4.7-flash 31.7% Jan 2026 Official leaderboard
7 GPT-4o (snapshot unspecified) 31.1% Official leaderboard
8 Meta-LLaMA-3-8B-Instruct 7.20% Official leaderboard
#ModelScoreSettingDateSource
1 Gemini 3 Flash 57.1% Dec 2025 Official leaderboard
2 Gemini 2.5 Flash Sep 2025 48.3% Sep 2025 Official leaderboard
3 Gemma 4 31B IT 46.4% Apr 2026 Official leaderboard
4 GPT-5.2 42.8% Dec 2025 Official leaderboard
5 Qwen3-32B 42.6% Apr 2025 Official leaderboard
6 glm-4.7-flash 31.7% Jan 2026 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Own leaderboard Not tested since May 2026

Source: Official K12-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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Flash Gemini 2.5 Flash Qwen3-32B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Flash Gemini 2.5 Flash

Source: Official K12-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 Knowledge

All 332 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy. Seed1.8 92.3%
  • SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (8 entries).
  • Pass Models scored 6 models we track, plus 2 other entries on its leaderboard.
  • Warning Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.2 (#51).
  • Pass Headroom The best is 57.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on raw.githubusercontent.com: 6 rows for models we track and 2 other entries at the last read.
  • Pass Last read November 19th, 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