Language & multilingual benchmarks
Explore language understanding, translation, and multilingual tasks. Compare coverage across languages and check whether an evaluation measures fluency, comprehension, or reasoning in a particular language.
Published
Updated:
Models tested: 15
Motivated by this gap, we developed the Multilingual Native Reasoning Challenge (MultiNRC), a new multilingual evaluation benchmark that contains native and challenging reasoning…
Top models · scale
Latest result per model · Version not reported
- 1Muse Spark 1.165.59 source score
- 2Gemini 3.1 Pro Preview64.74 source score
- 3Claude Opus 4.6Non-thinking48.34 source score
Updated:
Models tested: 8
The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset…
Top models · scale
Latest result per model · Version not reported
- 1GPT-4 Turbo Preview60.76 source score
- 2Gemini 2.0 Flash55.6 source score
- 3Claude 3 Sonnet54.17 source score
Updated:
Models tested: 5
The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset…
Top models · scale
Latest result per model · Version not reported
- 1O3 Mini1,156 source score
- 2DeepSeek R11,052 source score
- 3Gemini 1.5 Flash1,015 source score
Updated:
Models tested: 5
The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset…
Top models · scale
Latest result per model · Version not reported
- 1O3 Mini1,079 source score
- 2GPT-4 Turbo Preview1,020 source score
- 3Gemini 1.5 Flash980 source score
Updated:
Models tested: 4
The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset…
Top models · scale
Latest result per model · Version not reported
- 1O3 Mini1,104 source score
- 2Gemini 1.5 Flash966 source score
- 3Gemma 2 27B892 source score
Updated:
Models tested: 4
The Scale AI Multilingual Prompts Dataset, composed of 1,000 prompts per language, is tailored to enhance models’ interaction capabilities across multiple languages. This dataset…
Top models · scale
Latest result per model · Version not reported
- 1O3 Mini1,093 source score
- 2Gemini 1.5 Flash967 source score
- 3Llama 3.3 70B Instruct808 source score
Updated:
Models tested: 151
Human pairwise preferences on english prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,537.62
- 2Claude Opus 5.5High1,519.03
- 3Claude Opus 4.6High1,511.23
Updated:
Models tested: 151
Human pairwise preferences on non english prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,521.53
- 2Claude Fable 5.1Max1,508.06
- 3Claude Opus 5.5High1,507.23
Updated:
Models tested: 151
Human pairwise preferences on english prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,531.99
- 2Muse WonderMax1,516.57
- 3Claude Opus 5.5High1,513.29
Updated:
Models tested: 151
Human pairwise preferences on non english prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,510.75
- 2Muse WonderMax1,498.84
- 3Claude Opus 5.5High1,497.7
Updated:
Models tested: 146
Human pairwise preferences on chinese prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,611.06
- 2Claude Fable 5.1Max1,583.86
- 3Claude Opus 5.5High1,582.32
Updated:
Models tested: 146
Human pairwise preferences on chinese prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,588.95
- 2Claude Opus 5.5High1,571.44
- 3Claude Fable 5.1Max1,568.05
Updated:
Models tested: 150
Human pairwise preferences on russian prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 4 ArgonHigh1,533.75
- 2Muse WonderMax1,522.24
- 3Claude Fable 5.1Max1,519.55
Updated:
Models tested: 150
Human pairwise preferences on russian prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Muse WonderMax1,533
- 2Gemini 4 ArgonHigh1,525.63
- 3Gemini 3.7 FlashHigh1,518.23
Updated:
Models tested: 132
Human pairwise preferences on french prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Muse WonderMax1,523.45
- 2Claude Fable 5.1Max1,523.26
- 3Claude Opus 5High1,518.79
Updated:
Models tested: 132
Human pairwise preferences on french prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Muse WonderMax1,535.35
- 2Claude Fable 5High1,523.66
- 3Muse Spark 1.3Max1,520.44
Updated:
Models tested: 134
Human pairwise preferences on spanish prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Opus 5Max1,523.42
- 2Claude Fable 5.1Max1,520.31
- 3Claude Opus 4.61,509.6
Updated:
Models tested: 134
Human pairwise preferences on spanish prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5High1,504.53
- 2Claude Fable 5.1Max1,500.33
- 3Claude Opus 4.61,498.93
Updated:
Models tested: 120
Human pairwise preferences on japanese prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5.1Max1,562.18
- 2Claude Opus 5.5High1,533.98
- 3Claude Opus 5Max1,520.11
Updated:
Models tested: 120
Human pairwise preferences on japanese prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5.1Max1,553.51
- 2Claude Opus 5.5High1,530.97
- 3Claude Fable 5High1,518.1
Updated:
Models tested: 124
Human pairwise preferences on korean prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Opus 5Max1,517.21
- 2Claude Fable 5.1Max1,516.7
- 3Claude Opus 5High1,510.96
Updated:
Models tested: 124
Human pairwise preferences on korean prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5.1Max1,496.84
- 2Claude Opus 5Max1,495.38
- 3Claude Fable 5High1,495.34
Updated:
Models tested: 129
Human pairwise preferences on german prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Opus 5Max1,527.46
- 2Gemini 3 Pro1,515.24
- 3Muse Spark 1.3Max1,512.58
Updated:
Models tested: 129
Human pairwise preferences on german prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Gemini 3 Pro1,520.69
- 2Muse Spark 1.3Max1,514.84
- 3Claude Opus 5Max1,509.6
Updated:
Models tested: 117
Human pairwise preferences on polish prompts in Arena’s text evaluation. The original organizer publishes model ratings and uncertainty; these are preference ratings rather than…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5.1Max1,525.09
- 2Claude Opus 5Max1,521.86
- 3Claude Opus 5High1,516.76
Updated:
Models tested: 117
Human pairwise preferences on polish prompts in Arena’s text style control evaluation. The original organizer publishes model ratings and uncertainty, adjusting for response…
Top models · Arena
Latest result per model · Version not reported
- 1Claude Fable 5.1Max1,514.16
- 2Claude Fable 5High1,513.25
- 3Muse Spark 1.11,510.83
Methodology and provenance
Benchize collects and structures public benchmark reports. Scores, costs, timings, ranks and uncertainty remain attributed to their original sources. Different versions, configurations and measurement units are separate results. Missing fields stay unavailable.