Examenos

64.39 points · raw 64.39 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: overall.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 64.39.

94.12 points · raw 94.12 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlg.translation-en-xx.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 94.12.

93.08 points · raw 93.08 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlg.translation-xx-en.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 93.08.

16.90 points · raw 16.9 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlg.abssum.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 16.9.

55.25 points · raw 55.25 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlg.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 55.25.

74.65 points · raw 74.65 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlr.nli.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 74.65.

89.80 points · raw 89.8 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlr.causal.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 89.8.

82.23 points · raw 82.23 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlr.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 82.23.

71.24 points · raw 71.24 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlu.metaphor.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 71.24.

81.93 points · raw 81.93 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlu.qa.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 81.93.

74.02 points · raw 74.02 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlu.sentiment.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 74.02.

75.73 points · raw 75.73 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.nlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 75.73.

90.48 points · raw 90.48 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.instruction-following.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 90.48.

53.75 points · raw 53.75 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.multi-turn.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 53.75.

49.39 points · raw 49.39 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.safety.toxicity.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 49.39.

49.39 points · raw 49.39 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.safety.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 49.39.

32.26 points · raw 32.26 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.linguistic-diagnostics.syntax.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 32.26.

66.77 points · raw 66.77 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.linguistic-diagnostics.pragmatics.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 66.77.

49.52 points · raw 49.52 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.linguistic-diagnostics.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 49.52.

65.19 points · raw 65.19 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: id.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 65.19.

87.39 points · raw 87.39 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlg.translation-id-xx.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 87.39.

91.15 points · raw 91.15 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlg.translation-xx-id.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 91.15.

89.27 points · raw 89.27 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlg.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 89.27.

59.29 points · raw 59.29 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlu.metaphor.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 59.29.

87.06 points · raw 87.06 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlu.qa-mc.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 87.06.

71.39 points · raw 71.39 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlu.sentiment.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 71.39.

72.58 points · raw 72.58 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.nlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 72.58.

73.33 points · raw 73.33 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.instruction-following.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 73.33.

49.18 points · raw 49.18 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.multi-turn.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 49.18.

71.09 points · raw 71.09 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: jv.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 71.09.

82.75 points · raw 82.75 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlg.translation-id-xx.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 82.75.

89.89 points · raw 89.89 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlg.translation-xx-id.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 89.89.

86.32 points · raw 86.32 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlg.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 86.32.

48.09 points · raw 48.09 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlu.metaphor.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 48.09.

79.05 points · raw 79.05 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlu.qa-mc.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 79.05.

71.22 points · raw 71.22 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlu.sentiment.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 71.22.

66.12 points · raw 66.12 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.nlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 66.12.

72.63 points · raw 72.63 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.instruction-following.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 72.63.

51.01 points · raw 51.01 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.multi-turn.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 51.01.

69.02 points · raw 69.02 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: su.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 69.02.

77.08 points · raw 77.08 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.nlg.translation-id-xx.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 77.08.

81.31 points · raw 81.31 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.nlg.translation-xx-id.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 81.31.

79.20 points · raw 79.2 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.nlg.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 79.2.

66.13 points · raw 66.13 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.nlu.sentiment.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 66.13.

66.13 points · raw 66.13 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.nlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 66.13.

68.79 points · raw 68.79 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.instruction-following.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 68.79.

42.23 points · raw 42.23 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.multi-turn.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 42.23.

64.09 points · raw 64.09 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: ban.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 64.09.

54.75 points · raw 54.75 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlg.translation-id-xx.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 54.75.

73.67 points · raw 73.67 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlg.translation-xx-id.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 73.67.

64.21 points · raw 64.21 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlg.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 64.21.

54.13 points · raw 54.13 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlu.qa.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 54.13.

56.66 points · raw 56.66 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlu.sentiment.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 56.66.

55.40 points · raw 55.4 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.nlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 55.4.

47.43 points · raw 47.43 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.instruction-following.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 47.43.

43.18 points · raw 43.18 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.multi-turn.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 43.18.

52.55 points · raw 52.55 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: bbc.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 52.55.

58.48 points · raw 58.48 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 58.48.

64.14 points · raw 64.14 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.Humanities.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 64.14.

62.44 points · raw 62.44 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.Indonesian language.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 62.44.

39.83 points · raw 39.83 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.Local languages and cultures.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 39.83.

65.17 points · raw 65.17 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.Social science.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 65.17.

60.82 points · raw 60.82 points

Headline · unknown effort · Own vendor

Source/record date: 2026-10-18

First-party GoTo-AI leaderboard snapshot updated 18 October 2026. SEA-HELM/IndoMMLU localized tasks; native 0–100 score points. Multiple-choice normalization removes the random-chance baseline, so these are not raw accuracy values. Source describes 8 runs and 30 bootstrap resamples and means of competency/task scores. Card specifies native zero-shot prompts and 100–1000 instances per dataset. SEA-IFEval additionally requires correct language; SEA-MTBench is GPT-4-1106-preview judging two-turn answers against GPT-3.5-turbo-0125 with ties 0.5. Run date and reasoning effort unpublished. Exact source path: indommlu.STEM.total. Distinct language/task/category score; identical single-component/category numbers are different recorded source fields. Source score label: 60.82.