Full 90-Language Benchmark

March 17, 2026 ยท View on GitHub

This is a comprehensive multilingual evaluation covering 90 languages, comparing Chandra 2 against Gemini 2.5 Flash. The average scores are lower than the 43-language benchmark because this includes many lower-resource languages.

Overall Scores

Chandra 2Gemini 2.5 Flash
Average72.7% +/- 1.2%60.8% +/- 1.3%

Results by Language

LanguageChandra 2Gemini 2.5 Flash
af80.4%85.8%
am34.4%0.5%
ar68.4%84.4%
as35.8%23.1%
az75.2%74.0%
be80.7%66.4%
bg83.1%64.3%
bn72.8%55.3%
br90.0%69.4%
bs84.8%85.1%
ca85.1%88.0%
cs85.3%79.1%
cy82.2%77.6%
da91.1%86.0%
de94.8%88.3%
el85.6%83.5%
en96.6%90.3%
eo80.1%71.9%
es89.3%86.8%
et75.2%73.7%
eu80.2%74.6%
fa75.1%61.8%
fi83.4%86.0%
fr93.7%86.1%
fy81.2%70.1%
ga80.9%70.1%
gd71.8%59.5%
gl80.9%80.9%
gu70.8%47.6%
ha72.1%59.1%
he70.4%50.9%
hi78.4%82.7%
hr90.1%88.2%
hu82.1%84.5%
hy64.2%42.1%
id91.6%88.3%
is77.3%72.2%
it94.6%85.7%
ja86.9%80.0%
jv73.2%80.4%
ka77.0%39.3%
kk80.5%77.2%
km46.1%6.3%
kn63.2%24.5%
ko81.5%84.8%
ku62.0%63.2%
ky81.2%69.8%
la73.8%70.5%
lo60.9%13.3%
lt79.8%70.5%
lv76.9%81.5%
mg81.2%78.4%
mk83.5%77.4%
ml64.3%23.8%
mn88.4%71.4%
mr75.0%69.7%
ms79.3%79.8%
my55.9%15.8%
ne45.3%43.0%
nl88.6%87.5%
no90.5%87.8%
or31.1%11.2%
pa48.3%22.4%
pl91.5%91.1%
ps12.6%13.3%
pt95.2%89.4%
ro84.5%76.7%
ru85.5%82.8%
sa51.1%44.6%
sd50.0%29.3%
si62.4%26.2%
sk77.3%81.2%
sl81.0%80.1%
so82.4%69.9%
sq75.3%77.1%
sr90.3%89.7%
su85.7%96.4%
sv93.3%91.1%
sw88.9%80.9%
ta77.7%53.9%
te58.6%33.3%
th62.6%66.7%
tr84.1%84.1%
ug25.8%5.4%
uk91.0%87.9%
ur44.1%57.6%
uz77.2%52.8%
vi82.6%89.5%
xh82.1%62.1%
yi24.9%6.8%
zh88.7%70.0%