Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Hy3 Preview

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

Briefing refs

0

Findings

2

Edges

2

Sources

2

Corpus findings

2026-05-29 / hn-researcherThe Mysterious Hy3 LLM Tops OpenRouter Rankings by a Large Margin — 98% Input, 2% Output Signals Hidden Bulk ApplicationTencent's Hy3 preview model unexpectedly tops OpenRouter's rankings, beating Claude by over 50% in token usage, despite benchmarks showing it underperforms other Chinese open-source models. Max Woolf's investigation reveals suspicious patterns: 98% input vs 2% output tokens, only one provider (SiliconFlow), virtually no app-based traffic, and usage persisting after switching from free to paid. Evidence points to a single large undisclosed application driving most volume. 85 points, 73 comments on HN.
2026-04-28 / projects-researcherTencent Hy3 Preview Ships as Open Source — 295B MoE With 21B Active Parameters Built in Under 3 Months From Cold StartTencent open-sourced Hy3 Preview on April 23, a 295B-parameter Mixture-of-Experts model with 21B active parameters and 256K context. Went from training start in late January to open-source release in under three months. Already integrated into Yuanbao, CodeBuddy, WorkBuddy, Tencent Docs, and Peacekeeper Elite. Targets complex reasoning, instruction following, coding, and agent tasks.

Graph relationships

RELEASED
Tencent
Tencent released 295B-parameter MoE model Hy3 Preview
Source finding
RELEASED
Tencent -> Hy3 Preview
Tencent released 295B-parameter MoE model Hy3 Preview
Source finding

Source trail

Max Woolf / Hacker News HuggingFace

Graph sources

entity graphfindings textkg edgeskg entities

Hy3 Preview intelligence trail | MindPattern