Tencent releases Hy4 open-weight AI model with visual understanding and filtered cyber data
Tencent released Hy4, a 770-billion-parameter open-weight model that adds image understanding for the first time. The company says it limited hacking-related training data to curb offensive cyber capabilities.
Updated (version 2). New coverage since the last version from Anadolu Agency, The Straits Times.
1 / 4
The story, neutrally told
Centre · 2Tencent released its new flagship AI model, Hy4, on Wednesday 4 November 2026, describing it as an open-weight model. South China Morning PostN “Released on Wednesday, the large language model (LLM) introduced visual understanding for the first time.” Read at South China Morning Post ↗ CNAN “Tencent on Wednesday released Hy4, its new open-source model” Read at CNA ↗ Centre · 1The Hy4 series has 770 billion parameters, a 2.6-fold increase from the 295 billion in its predecessor, Hy3, and for the first time it can process images, analyse complex charts and do visual mathematical reasoning, according to Tencent. South China Morning PostN “The Hy4 series features 770 billion parameters – a 2.6-fold jump from the 295 billion parameters in its predecessor, Hy3.”“Hy4 could process images, analyse complex charts and conduct visual mathematical reasoning, according to the Shenzhen-based company.” Read at South China Morning Post ↗ Centre · 1Tencent said Hy4 is now "one of the strongest open-weight models available today" but acknowledged it still trails the leading closed-source systems on hard internal benchmarks such as research-level mathematics, long-horizon coding, model self-improvement and advanced visual question answering. South China Morning PostN “although it acknowledged that the model still trailed the world’s leading closed-source systems” Read at South China Morning Post ↗
Centre · 1According to benchmarks in Tencent's technical blog, Hy4 outperformed Z.ai's GLM-5.3 and DeepSeek-V4.1-Flash-1017 on all tested tasks and traded wins with Moonshot's Kimi K3.1 Pro; these are the company's own figures. South China Morning PostN “The new flagship outperformed Z.ai’s GLM-5.3 and the DeepSeek-V4.1-Flash-1017 model across all tested tasks, while trading wins with Moonshot’s Kimi K3.1 Pro” Read at South China Morning Post ↗ Centre · 1Tencent said Hy4 powers an update to its workplace AI tool WorkBuddy, scoring 83.18 on average with a 96.15 per cent pass rate in internal evaluations. It also said a lightweight Hy4 Flash is coming soon and that a next-generation Hy5 will scale further. South China Morning PostN “The model achieved an average score of 83.18 and a pass rate of 96.15 per cent in internal evaluations for WorkBuddy tasks”“Tencent indicated that its next-generation model, Hy5, would scale further, while a soon-to-come, lightweight Hy4 Flash version would slash task costs.” Read at South China Morning Post ↗ Centre · 2Tencent said it filtered out training content on exploit development, reverse engineering and penetration testing to limit Hy4's potential offensive cyber capabilities. CNAN “Tencent said it filtered out content during training such as exploit development, reverse engineering, and penetration testing to limit Hy4's potential "offensive" capabilities.” Read at CNA ↗ South China Morning PostN ““to avoid amplifying potential offensive capabilities”” Read at South China Morning Post ↗
Centre · 1CNA links this to a slew of AI-driven cyberattacks that raised concerns about advanced models, and notes Moonshot AI limited its newest model's cyber-offensive capabilities last month. Tencent said it is "committed to a more responsible approach to model training, open-sourcing, and release". CNAN “Moonshot AI last month limited its newest model's cyber-offensive capabilities, as Chinese AI firms join US companies to prevent loss-of-control risks.”“we are committed to a more responsible approach to model training, open-sourcing, and release” Read at CNA ↗ Centre · 1The launch comes as Tencent spends heavily on AI: its capital expenditure rose 176 per cent year on year to 52.8 billion yuan (US$7.9 billion) in the second quarter, pushing free cash flow negative. A preview of Hy4 released in late August ranked 23rd on Code Arena's WebDev leaderboard on Wednesday. The South China Morning Post noted that its owner, Alibaba, is one of Tencent's domestic rivals. South China Morning PostN “increased its capital expenditure by 176 per cent year on year to 52.8 billion yuan (US$7.9 billion) in the second quarter”“A preview version of Hy4, launched in late August, ranked 23rd globally on Wednesday on Code Arena’s WebDev leaderboard”“Alibaba owns the South China Morning Post.” Read at South China Morning Post ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left0 outlets
No left outlet in our sources has covered this story yet.
Centre3 outlets
- Framing
- SCMP treats the release as a competitive move in China's open-weight AI race, leading on capabilities and benchmarks; CNA (Reuters text) leads on the cybersecurity-safety angle.
- Emphasis
- SCMP: parameter count, benchmarks against DeepSeek, Z.ai, Moonshot, capex. CNA: limited cyber training data and AI-driven hacking concerns.
- Leaves out or plays down
- CNA omits the visual understanding, model size and benchmark detail; SCMP gives the cyber filtering only a closing line and no wider context on attacks.
- For example
-
“seeking to challenge top open-weight rivals with enhanced agentic capabilities built to power its enterprise software ecosystem.” — South China Morning Post
“The rollout comes after a slew of AI-driven cyberattacks around the world raised concerns about the capabilities of advanced models” — CNA
Right1 outlet
- Framing
- The Straits Times ran the Reuters cybersecurity-focused report unchanged, framing the launch as a safety response to AI-driven hacking fears.
- Emphasis
- Tencent's filtering of offensive-security training content, the Moonshot AI precedent and Tencent's responsibility statement.
- Leaves out or plays down
- Leaves out visual understanding, the 770-billion-parameter size, benchmark comparisons and Tencent's AI spending.
- For example
-
“Tencent said it filtered out content during training such as exploit development, reverse engineering, and penetration testing to limit Hy4‘s potential “offensive” capabilities.” — The Straits Times
What every side reports
- Tencent released Hy4 on Wednesday 4 November 2026.
- Tencent filtered exploit development, reverse engineering and penetration-testing content from training to limit offensive cyber capabilities.
-
Tencent filtered exploit development, reverse engineering and penetration testing content from Hy4's training to limit offensive cyber capabilities.
Reported- Reports 3
- CNA, South China Morning Post, The Straits Times
-
Tencent released its new flagship model Hy4 on Wednesday 4 November 2026 as an open model.
Reported- Reports 3
- CNA, South China Morning Post, The Straits Times
Tencent organisation
Tencent presents Hy4 as one of the strongest open-weight models, admits a gap to top closed-source systems, and says it is taking a responsible approach to cyber and biological risks.
““We will continue to steadily scale pre-training and post-training to narrow this gap,”” — South China Morning Post
“We recognize the potential safety risks posed by advanced cyber and biological capabilities” — CNA
DeepSeek organisation
DeepSeek is named as a rival; Tencent's benchmarks say Hy4 beat DeepSeek-V4.1-Flash-1017 on all tested tasks. DeepSeek is not quoted.
“The new flagship outperformed Z.ai’s GLM-5.3 and the DeepSeek-V4.1-Flash-1017 model across all tested tasks” — South China Morning Post
Left0 articles
No coverage yet.
Centre3 articles
-
Tencent unveils Hy4 as open-weight AI race with DeepSeek, Z.ai, Alibaba heats up
Neutral Business-tech report on Hy4 as Tencent's push in the open-weight race, relaying company benchmarks.
-
CNAN ·
Tencent rolls out AI model trained with limited cybersecurity data amid hacking fears
Neutral Short report framing Hy4 around cybersecurity precautions amid AI-driven hacking fears.
-
Right1 article
-
The Straits TimesRC · · via Reuters
Tencent rolls out AI model trained with limited cybersecurity data amid hacking fears
Neutral Reuters wire copy carried by The Straits Times, centred on Tencent limiting offensive cyber capabilities amid concerns over AI-driven attacks.
- 4 Nov 07:01 First South China Morning PostN Tencent unveils Hy4 as open-weight AI race with DeepSeek, Z.ai, Alibaba heats up
- 4 Nov 10:37 +3h 36m CNAN Tencent rolls out AI model trained with limited cybersecurity data amid hacking fears
- 4 Nov 11:31 +4h 31m The Straits TimesRC Tencent rolls out AI model trained with limited cybersecurity data amid hacking fears via Reuters
- 4 Nov 20:25 +13h 24m Anadolu AgencyN Chinese tech giant Tencent launches new AI model with safety safeguards
Times are when each article was published, or when we first saw it if the outlet gave no time.
Ask about this story
Loading…