ModelChorusModelChorus
Ịma akaNkataLeaderboardSụgharịaNtuleAkụkọ
Otu o si arụ ọrụAPIUsoro ỌrụIwu Nzuzo

Ikike nchekwa 2026 MeetKai Inc.

Ntule

Akara akpaka nke lm-eval-harness n'ofe ọrụ ọtụtụ asụsụ. Ihe dị elu ka mma — pịa ahịrị ma ọ bụ mkpụrụ iji banye.

N'otu nlele

Asụsụ ndị emechiri
12
Asụsụ kacha mma
Albanian (Albania)
86.1%
Asụsụ kacha sie ike
Yoruba (Nigeria)
46.7%
Mkpuchi nyocha
13%
182 n'ime 1407 mkpụrụ

Asụsụ site na isi ike

Avọọl akara n'ofe nyocha niile

  • Yoruba (Nigeria)
    47
  • Spanish (Spain)
    47
  • Igbo (Nigeria)
    48
  • Hausa (Nigeria)
    51
  • Arabic (Saudi Arabia)
    53
  • Urdu (Pakistan)
    57
  • English (US)
    60
  • French (France)
    65
  • Portuguese (Portugal)
    76
  • Swahili (Tanzania)
    79
  • Ukrainian (Ukraine)
    82
  • Albanian (Albania)
    86

Nyocha kacha mma maka asụsụ ọ bụla

Onye-isi site na avọọl akara (opekata 3 nyocha)

  • Arabic (Saudi Arabia)
    Functionary Swahili Large
    69.4%
  • English (US)
    Functionary Swahili Large
    83.7%
  • Spanish (Spain)
    Functionary Swahili Large
    62.8%
  • French (France)
    GLM 5
    84.9%
  • Hausa (Nigeria)
    GPT-5 Nano
    55.0%
  • Portuguese (Portugal)
    GPT-5 Nano
    78.0%
  • Albanian (Albania)
    GPT-oss-120B
    87.3%
  • Swahili (Tanzania)
    Functionary Swahili Large
    89.9%
  • Ukrainian (Ukraine)
    Functionary Swahili Mini
    91.8%
  • Urdu (Pakistan)
    Functionary Swahili Mini
    67.2%

Ọrụ ndị kachasị sie ike

Avọọl kacha ala n'etiti nyocha ndị tozuru oke

  • hausa afriqa
    Hausa (Nigeria)
    12.5%
  • urdu simpleqa
    Urdu (Pakistan)
    21.2%
  • english belebele
    English (US)
    27.4%
  • urdu emotion class
    Urdu (Pakistan)
    29.7%
  • spanish global mmlu
    Spanish (Spain)
    33.1%
  • arabic tydiqa
    Arabic (Saudi Arabia)
    33.7%
  • spanish belebele
    Spanish (Spain)
    35.0%
  • ifeval
    English (US)
    37.7%

Ọrụ ndị kachasị mfe

Avọọl kacha n'etiti nyocha ndị tozuru oke

  • ukrainian polywrite
    Ukrainian (Ukraine)
    92.9%
  • albanian belebele
    Albanian (Albania)
    91.1%
  • french sib200
    French (France)
    88.9%
  • ukrainian sib200
    Ukrainian (Ukraine)
    88.4%
  • swahili sib200
    Swahili (Tanzania)
    87.4%
  • arabic sib200
    Arabic (Saudi Arabia)
    86.3%
  • english mgsm
    English (US)
    86.2%
  • albanian sib200
    Albanian (Albania)
    85.9%

Nyocha niile

#ModelNkwkArabicEnglishSpanishFrenchHausaIgboPortugueseAlbanianSwahiliUkrainianUrduYorubaLee nkọwa zuru ezu
1
GPT-5 Nano
52/67
68.463.663.961.765.655.0--78.085.176.180.159.5--
2
Functionary Swahili Large
29/67
67.769.483.762.876.846.1--69.385.889.975.259.1--
3
Functionary Swahili Mini
22/67
62.243.423.6--54.350.8--75.885.485.991.867.2--
4
GPT-oss-120B
59.138.966.843.658.750.248.077.487.374.181.647.146.7
5
GLM 5
12/67
57.0--53.231.884.9--------79.4------

Isi na isi

Gosi

Site na asụsụ

Avọọl akara na ọrụ nyocha abụọ ahụ akpọrọ

  • Hausaọrụ 4
    GPT-5 Nano
    59.7
    Functionary Swahili Large
    46.1
  • Portugueseọrụ 3
    GPT-5 Nano
    76.7
    Functionary Swahili Large
    69.3
  • Arabicọrụ 2
    GPT-5 Nano
    62.6
    Functionary Swahili Large
    69.4
  • Spanishọrụ 1
    GPT-5 Nano
    67.7
    Functionary Swahili Large
    62.8
  • Frenchọrụ 3
    GPT-5 Nano
    72.7
    Functionary Swahili Large
    76.8
  • Swahiliọrụ 1
    GPT-5 Nano
    86.3
    Functionary Swahili Large
    89.9
  • Englishọrụ 2
    GPT-5 Nano
    87.2
    Functionary Swahili Large
    83.7
  • Ukrainianọrụ 2
    GPT-5 Nano
    73.8
    Functionary Swahili Large
    75.2
  • Albanianọrụ 3
    GPT-5 Nano
    87.0
    Functionary Swahili Large
    85.8
  • Urduọrụ 8
    GPT-5 Nano
    59.5
    Functionary Swahili Large
    59.1

Oghere maka ọrụ ọ bụla

Ọdịiche akara — mkpịsị aka nri = GPT-5 Nano n'ihu, mkpịsị aka aka ekpe = Functionary Swahili Large n'ihu

  • Hausa
    • hausa afrimgsm
      +24.8
    • hausa afrixnli
      +16.5
    • hausa afriqa
      +14.0
    • hausa sib200
      -1.1
  • Urdu
    • urdu freshqa
      +18.6
    • urdu fake news
      -13.8
    • urdu uquad
      +10.1
    • urdu facttool qa
      -7.5
    • urdu factcheckbench
      -7.1
    • urdu bingcheck
      +6.8
    • urdu emotion class
      -3.0
    • urdu simpleqa
      -1.0
  • Portuguese
    • portuguese hatebr
      +16.4
    • portuguese hate speech
      +6.5
    • portuguese tweetsentbr
      -0.7
  • Arabic
    • arabic tydiqa
      -10.6
    • arabic sib200
      -3.1
  • French
    • french fquad
      -8.3
    • french mgsm
      -2.8
    • french sib200
      -1.1
  • Albanian
    • albanian sib200
      +6.6
    • albanian belebele
      -2.9
    • albanian global mmlu
      -0.2
  • Ukrainian
    • ukrainian sib200
      -6.5
    • ukrainian squad
      +3.8
  • Spanish
    • spanish xquad es
      +5.0
  • English
    • english gsm8k
      +4.6
    • english mgsm
      +2.4
  • Swahili
    • swahili sib200
      -3.6

Mpaghara akara

GPT-5 Nano megide avọọl nke nyocha ndị ọzọ niile — mkpokọta ọrụ 52

  • Na-edu
    >+5pp
    21
  • Nhata
    ±5pp
    28
  • N'azụ
    -5 to -20pp
    3
  • N'azụ nke ukwuu
    <-20pp
    0

Mpaghara akara

Functionary Swahili Large megide avọọl nke nyocha ndị ọzọ niile — mkpokọta ọrụ 29

  • Na-edu
    >+5pp
    6
  • Nhata
    ±5pp
    15
  • N'azụ
    -5 to -20pp
    7
  • N'azụ nke ukwuu
    <-20pp
    1