Research

Paper

AI LLM February 25, 2026

IndicIFEval: A Benchmark for Verifiable Instruction-Following Evaluation in 14 Indic Languages

Authors

Thanmay Jayakumar, Mohammed Safi Ur Rahman Khan, Raj Dabre, Ratish Puduppully, Anoop Kunchukuttan

Abstract

Instruction-following benchmarks remain predominantly English-centric, leaving a critical evaluation gap for the hundreds of millions of Indic language speakers. We introduce IndicIFEval, a benchmark evaluating constrained generation of LLMs across 14 Indic languages using automatically verifiable, rule-based instructions. It comprises around 800 human-verified examples per language spread across two complementary subsets: IndicIFEval-Ground, translated prompts from IFEval (Zhou et al., 2023) carefully localized for Indic contexts, and IndicIFEval-Ground, synthetically generated instructions grounded in native Indic content. We conduct a comprehensive evaluation of major open-weight and proprietary models spanning both reasoning and non-reasoning models. While models maintain strong adherence to formatting constraints, they struggle significantly with lexical and cross-lingual tasks -- and despite progress in high-resource languages, instruction-following across the broader Indic family lags significantly behind English. We release IndicIFEval and its evaluation scripts to support progress on multilingual constrained generation (http://github.com/ai4bharat/IndicIFEval).

Metadata

arXiv ID: 2602.22125
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-02-25
Fetched: 2026-02-26 05:00

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.22125v1</id>\n    <title>IndicIFEval: A Benchmark for Verifiable Instruction-Following Evaluation in 14 Indic Languages</title>\n    <updated>2026-02-25T17:12:37Z</updated>\n    <link href='https://arxiv.org/abs/2602.22125v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.22125v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Instruction-following benchmarks remain predominantly English-centric, leaving a critical evaluation gap for the hundreds of millions of Indic language speakers. We introduce IndicIFEval, a benchmark evaluating constrained generation of LLMs across 14 Indic languages using automatically verifiable, rule-based instructions. It comprises around 800 human-verified examples per language spread across two complementary subsets: IndicIFEval-Ground, translated prompts from IFEval (Zhou et al., 2023) carefully localized for Indic contexts, and IndicIFEval-Ground, synthetically generated instructions grounded in native Indic content. We conduct a comprehensive evaluation of major open-weight and proprietary models spanning both reasoning and non-reasoning models. While models maintain strong adherence to formatting constraints, they struggle significantly with lexical and cross-lingual tasks -- and despite progress in high-resource languages, instruction-following across the broader Indic family lags significantly behind English. We release IndicIFEval and its evaluation scripts to support progress on multilingual constrained generation (http://github.com/ai4bharat/IndicIFEval).</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-02-25T17:12:37Z</published>\n    <arxiv:comment>8 pages + Appendix</arxiv:comment>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Thanmay Jayakumar</name>\n    </author>\n    <author>\n      <name>Mohammed Safi Ur Rahman Khan</name>\n    </author>\n    <author>\n      <name>Raj Dabre</name>\n    </author>\n    <author>\n      <name>Ratish Puduppully</name>\n    </author>\n    <author>\n      <name>Anoop Kunchukuttan</name>\n    </author>\n  </entry>"
}