Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Tower-Plus-72B

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

0

Findings

1

Edges

2

Sources

1

Corpus findings

2026-05-01 / sources-researcherTencent Open-Sources HY-MT1.5: 440MB Offline Translation Model That Outperforms Google Translate Across 33 LanguagesTencent Hunyuan Team open-sourced HY-MT1.5-1.8B-1.25bit, a 440MB on-device translation model supporting 33 languages and 1,056 translation directions. Uses Sherry 1.25bit quantization (accepted at ACL 2026) with 3:4 fine-grained sparsity to compress from 3.3GB FP16 to 440MB with minimal accuracy loss. Outperforms Tower-Plus-72B, Qwen3-32B, and commercial APIs including Microsoft Translator. Runs fully offline on phones via custom STQ mobile CPU kernel — no subscription, no internet required.

Graph relationships

BENCHMARKED AGAINST
HY-MT1.5
HY-MT1.5 outperforms Tower-Plus-72B on translation.
Source finding
BENCHMARKED AGAINST
HY-MT1.5 -> Tower-Plus-72B
HY-MT1.5 outperforms Tower-Plus-72B on translation.
Source finding

Source trail

Tencent / GitHub

Graph sources

entity graphfindings textkg edgeskg entities

Tower-Plus-72B intelligence trail | MindPattern