# Tencent's Hy4 preview raises coding benchmark position

_Thursday, September 3, 2026 at 4:30 AM EDT · AI · Latest · Tier 2 — Notable_

![Tencent's Hy4 preview raises coding benchmark position — Primary](https://www.silicon.co.uk/wp-content/uploads/2022/06/tencent-china-04.jpg)

Tencent's Hy4 preview AI model ranked eighth on Code Arena's WebDev leaderboard on Tuesday, compared with 34th for the prior Hy3 generation.

The 770-billion-parameter model has a one-million-token context window, up from 295 billion parameters and 256,000 tokens for Hy3. On the DeepSWE software-engineering benchmark, Hy4 scored 64.3, ahead of the cited scores for Alibaba's Qwen-3.8 Max and DeepSeek-V4 Pro. Analysts said Tencent deploys preview models across its ecosystem and uses resulting user data to improve them before full release.

## Sources

- [Silicon UK](https://www.silicon.co.uk/ai-2/tencent-ai-model-631347)

---
Canonical: https://techandbusiness.org/newswire/7ycjty_q_DBOeMWy_T76A0
Retrieved: 2026-09-03T11:25:18.684Z
Publisher: Tech & Business (techandbusiness.org)