Twitxy Saved

LILT

Verified account This account is verified. Learn more

@LILTLabs

We build the multilingual layer for English-first AI. Custom evals, benchmarks, and RL environments across 200+ languages. #BeyondTranslation
San Francisco Joined April 2026
18 Following 17 Followers 51 Posts

Latest posts 5

LILT Verified account @LILTLabs
1/ AURORA deep dive: Multilingual MultiChallenge. Multi-turn conversations testing instruction following, memory and self-coherence in Arabic, German, Korean and English.

Gemini 3.8 Flash is #1 in every non-English language, at under 1/6 the per-task cost of Opus 5.5 (#2).
LILT Verified account @LILTLabs
1/
AURORA deep dive: Multilingual τ³-bench. Multi-turn customer support with tool use (airline, retail, telecom, banking) in German, Korean, Turkish and Persian.

Claude Opus 5.5 leads GPT-6 Sol by ~9pp, at $0.28 per task vs $0.59 for Opus 5.
LILT Verified account @LILTLabs
1/
AURORA - Multilingual AI leaderboard deep dive: Multilingual Terminal-Bench. 324 coding tasks across 10 languages, designed and verified by native-language domain experts representing real challenges in software written for native-language users.

Claude Opus 5.5 takes #1, ~3pp ahead of Opus 5, with the biggest gains in Chinese and Korean. It's also ~45% cheaper per task: 20% lower price, ~30% fewer tokens.
LILT Verified account @LILTLabs
Announcing AURORA: LILT's multilingual AI leaderboard that measures frontier models on non-English agentic tasks grounded in language and culture.

Coding, customer support, instruction following, tool use. The rankings change with the language. 🧵
0:22
LILT Verified account @LILTLabs
1/ New benchmark: Terminal-Bench-LILT. 300 coding tasks, 10 languages, written by native-speaking authors, deterministic verifiers.

Best frontier model (GPT-5.5) resolves 63.1%. Many tasks: solved by zero of seven models. 🧵