Qwen releases BabelArena, a 23-language agent benchmark; all tested models score lower in low-resource languages
BabelArena is a multilingual agent benchmark covering 23 languages and four agent benchmark families across 13 domains, with 702 canonical tasks and 16,146 task instances. This release covers the standard subset for three families (τ³-Bench, VitaBench, DeepPlanning), totaling 532 tasks per language and 12,236 instances. Every component an agent touches is localized, including prompts, tasks, tool descriptions and environments. Five frontier models are reported, and all score lower in low-resource languages than in high-resource ones on Pass^1, Pass^3 and language consistency.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · concern recorded
- Discovery linkX
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Original repository+40
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
BabelArena is a multilingual agent benchmark covering 23 languages and four agent benchmark families across 13 domains, with 702 canonical tasks and 16,146 task instances. This release covers the standard subset for three families (τ³-Bench, VitaBench, DeepPlanning), totaling 532 tasks per language and 12,236 instances. Every component an agent touches is localized, including prompts, tasks, tool descriptions and environments. Five frontier models are reported, and all score lower in low-resource languages than in high-resource ones on Pass^1, Pass^3 and language consistency.
Key findings from the stored brief
- Benchmark covers 23 languages across high, medium and low resource levels, with identical task IDs in every language.
- All five evaluated frontier models score lower on Pass^1, Pass^3 and LC in low-resource languages than in high-resource ones.
- No single model leads every benchmark family, per the reported results.
- Translation quality averages 4.28, 4.02 and 3.91 out of 5 for high-, medium- and low-resource languages on 4,600 sampled translations.
The release covers only the standard subset for three of four families, and ClawEval task data is distributed separately; source claims not independently checked.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
The benchmark measures multilingual agent performance with fully localized environments and separate language-consistency scoring, exposing a performance gap for low-resource languages that translated-prompt-only benchmarks may not reveal.
Limits and open questions
The release covers only the standard subset for three of four families, and ClawEval task data is distributed separately; source claims not independently checked.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.