On Nov. 4, 2026, Tencent's Hy team published “Introducing Hy4,” the successor to Hy4 preview. Tencent calls it “another major leap forward that places Hy among the strongest open-source models for the first time,” though the weights are not yet released. The model is live on Tencent Cloud and OpenRouter and inside Tencent products, but its weights are not yet downloadable.
Key facts
| Item | Detail |
|---|---|
| Released | Nov. 4, 2026 |
| By | Tencent Hy team |
| New | vision and computer use (CLI and GUI) |
| Availability | Tencent Cloud, OpenRouter, CodeBuddy, WorkBuddy, Yuanbao, ima |
| API price per 1M tokens | $0.834 input, $2.501 output, $0.042 cached input |
| Weights | planned, not yet released |
What Tencent released
Hy4 is the full release after Hy4 preview. Tencent says it substantially scales up both pre-training and post-training. It adds vision capabilities for the first time and delivers major gains in instruction following, long-context understanding, long-horizon tasks, deep research, and more. Tencent keeps API pricing unchanged from Hy4 preview while acknowledging a notable gap remains with top proprietary models.
Vision and computer use
Tencent says Hy4 understands fine-grained image details and complex charts. Hy4 also supports computer use for the first time, working across command-line interfaces (CLI) and graphical user interfaces (GUI). Training targeted specialized software such as Blender and CAD, plus multi-app long-horizon workflows. In a demo, Hy4 locates a roughly 2-meter rooftop component inside a 2.8-gigapixel aerial image—a task that ties directly to the vision story Tencent highlights on its research page.
Coding and game development
Tencent worked with developers at Tencent Interactive Entertainment Group (IEG) on training data and evaluations, and adapted Hy4 for Unreal Engine, Unity, and Godot. Compared with Hy4 preview, Hy4 improves by 19.1% on GameDev and 9.8% on GameLogic. That pitch sits alongside broader agentic coding competition from models such as DeepSeek Open-Sources a Super-Agent Coding Prototype, though Hy4 remains API-first until weights ship.
Finance and office work
Tencent evaluated Hy4 with 213 finance professionals, including 153 from financial institutions, on 579 real-world financial tasks. Tencent says Hy4 ranks first among open-weight models (Tencent's framing) on that study (Hy4 3.04; DeepSeek-V4.1-Flash-1017 2.97; Kimi K3.1 Pro 2.68; GLM-5.3 2.68). In Tencent's internal evaluations, the hallucination rate fell from 15.76% on Hy4 preview to 5.45%, and speculative errors from 23.1% to 5.6%—figures Tencent attributes to its own benchmarks, not independent audits.
For everyday office work, Hy4 reaches chat history, emails, documents, calendars, and meeting notes through WorkBuddy's integration with WeCom. WorkBuddy's product team reports an average score of 83.18 and a 96.15% pass rate in its evaluations, ranking #1 among Chinese models in 9 of 12 scenarios.
Science and AI research
On TenureTrail, Tencent's internal benchmark of 123 research tasks across seven disciplines, Tencent says Hy4 ranks first among open-weight models (Tencent's framing). Tencent is open-sourcing two models co-developed by Hy4 and its Hyra 2.0 harness: HyCell for single-cell representation and HyPDE for physical field modeling—distinct from the main Hy4 weights, which remain planned.
Safety and the open-weights plan
Tencent limited cybersecurity training data to defensive material and ran red-team audits with Tencent Xuanwu Lab and Tencent Keen Security Lab. On Hy-BioExperimentBench, Hy4 scored 30.4/100, well below Tencent's high-risk threshold. Tencent says it plans to open-source Hy4's weights and has partnered with the vLLM (Inferact Inc.) and SGLang communities for ecosystem support, with a system card to accompany the weight release. Until then, callers use hosted APIs only.
Tencent also notes a notable gap remains between it and top proprietary models, particularly on challenging internal benchmarks for research-level mathematics, long-horizon coding, model self-improvement, and advanced visual question answering.
Price and where to use it
Hy4 is now available on Tencent Cloud and OpenRouter, as well as in products including CodeBuddy, WorkBuddy, Yuanbao, and ima. It delivers major performance gains over Hy4 preview at the same API price: cached input $0.042, input $0.834, and output $2.501 per million tokens. Tencent plans to release Hy4 flash in the near future with better cost-efficiency and faster inference. Multimodal agents from vendors such as Meituan Launches LongCat-2.5, a 1.6-Trillion-Parameter Model show how crowded that tier is becoming.
What it means
Analysis (ours). Hy4 is Tencent's bid to keep pace with Western and Chinese frontier labs without raising API prices: vision and GUI/CLI computer use are the headline upgrades, while finance and WorkBuddy scores aim at enterprise buyers. The planned vLLM and SGLang partnerships signal that open weights matter for credibility even when the hosted model ships first. Treat Tencent's finance and TenureTrail rankings as vendor-reported until outsiders reproduce them.
What to watch
Whether Hy4 flash arrives on schedule, when weights and the system card drop, and independent benchmarks on computer-use tasks outside Tencent's demos.