ECLeKTic
Models answer 384 closed-book questions, translated into 12 languages, about facts found in only one language's Wikipedia. Scoring checks transfer across languages.
In the authors’ words
To achieve equitable performance across languages, large language models (LLMs) must be able to abstract knowledge beyond the language in which it was learnt. However, the current literature lacks reliable ways to measure LLMs' capability of such cross-lingual knowledge transfer. To that end, we present ECLeKTic, a multilingual closed-book QA dataset that Evaluates Cross-Lingual Knowledge Transfer in a simple, black-box manner. Concretely, we used the presence and absence of Wikipedia articles in 12 languages to detect pieces of information that were likely available during pre-training in one of the languages but not in the others. We curate ECLeKTic as a set of fact-seeking questions over this kind of information, in all the different languages. Therefore, in order to solve ECLeKTic the model is required to transfer knowledge between languages. We evaluated 8 LLMs and showed that current SOTA models struggle to effectively share knowledge across languages, even if they can predict the answer for questions in the language in which the knowledge was acquired.
- Level
- Everyday Most adults, without training
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Short answer A number, word or short string matched to a key.
Claude Mythos Preview leads with 56.3% among models released by May 2026
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in May 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 15 more
- 1 Claude Mythos Preview 56.3%
- 2
Gemini 2.5 Pro Jun 2025 46.8% - 3 Claude Opus 4.8 39%
- 4 Claude Opus 4.6 37.3%
- 5 Gemini 2.5 Flash Jun 2025 36.8%
- 6 Gemma 3 27B 24.4%
- 7 Claude Opus 4.7 24.1%
- 8 Gemma 3 12B 10.3%
- 9 Gemma 3 4B 4.60%
- 10 Gemma 3 1B 1.40%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Mythos Preview | 56.3% | Apr 2026 | Model card | |
| 2 | Gemini 2.5 Pro Jun 2025 | 46.8% | Jul 2025 | Model card | |
| 3 | Claude Opus 4.8 | 39% | May 2026 | Model card | |
| 4 | Claude Opus 4.6 | 37.3% | Feb 2026 | Model card | |
| 5 | Gemini 2.5 Flash Jun 2025 | 36.8% | Jul 2025 | Model card | |
| 6 | Gemma 3 27B | 24.4% | Mar 2025 | Model card | |
| 7 | Claude Opus 4.7 | 24.1% | Apr 2026 | Model card | |
| 8 | Gemma 3 12B | 10.3% | Mar 2025 | Model card | |
| 9 | Gemma 3 4B | 4.60% | Mar 2025 | Model card | |
| 10 | Gemma 3 1B | 1.40% | Mar 2025 | Model card |
The benchmark's own leaderboard has 35 models. We plot every source in the charts below.
2 models have held number one since its release in Feb 2025
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
No launch post has listed it since Jun 2025
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 26 Jun 2025 | Gemma 3n E4B | 1.9 | ECLeKTic score 0-shot (IT) |
| 12 Mar 2025 | Gemma 3 27B | 24.4 | ECLeKTic (PT); IT 0-shot = 16.7 |
10 system cards report it, from 2 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 28 May 2026 | Claude Opus 4.8 system card | 39 | ECLeKTic full cross-lingual set (multilingual factual questions), net score 0.39 (correct minus incorrect), closed book, printed on chart (6.3.3.1, Figure 6.3.3.1.B) | Hallucination |
| 16 Apr 2026 | Claude Opus 4.7 system card | 24.1 | ECLeKTic multilingual, full cross-lingual set, net score (correct minus incorrect), closed book, printed on chart. Correct 35.4, unsure 53.3, incorrect 11.3 (6.3.3.1, Figures 6.3.3.1.A and B) | Hallucination |
| 7 Apr 2026 | Claude Mythos Preview system card | 56.3 | Multilingual factual hallucinations, ECLeKTic full dataset including cross-lingual translations, correct rate, value printed on chart; unsure 31.5, incorrect 12.1 (4.3.3.2, Figure 4.3.3.2.A) | Hallucination |
| 5 Feb 2026 | Claude Opus 4.6 system card | 37.3 | ECLeKTic subset in original languages, net score (correct minus incorrect), extended thinking, value printed on chart; correct rate 53.9, net 34.6 with thinking plus effort (Figure 4.3.A) | Hallucination |
| 7 Jul 2025 | Gemini 2.5 Pro technical report | 46.8 | ECLeKTic (Table 3) | Capabilities |
| 7 Jul 2025 | Gemini 2.5 Flash technical report | 36.8 | ECLeKTic (Table 3) | Capabilities |
| 25 Mar 2025 | Gemma 3 4B technical report | 4.6 | ECLeKTic score, IT, 0-shot (Table 18). PT 2-shot (Table 13): 11.0 | Capabilities |
| 25 Mar 2025 | Gemma 3 27B technical report | 16.7 | ECLeKTic score, IT, 0-shot (Table 18). PT 2-shot (Table 13): 24.4 | Capabilities |
| 25 Mar 2025 | Gemma 3 1B technical report | 1.4 | ECLeKTic score, IT, 0-shot (Table 18). PT 2-shot (Table 13): 4.69 | Capabilities |
| 25 Mar 2025 | Gemma 3 12B technical report | 10.3 | ECLeKTic score, IT, 0-shot (Table 18). PT 2-shot (Table 13): 17.2 | Capabilities |
More in Multilingual
- MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
- WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET.
K-EXAONE 90.5% - MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language.
Qwen3.7-Max 87% - CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories.
Solar Open 2 90.7% - Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy.
Inkling 88.7% - MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%