Tencent’s Hy4 AI Beats DeepSeek in Blind Test by 213 Finance Pros

Tencent’s new Hy4 AI model beat open-source rivals in a finance expert blind test while keeping prices flat and adding free access offers.

maisiekooc
Maisie Morrison

AgentLocker Editor

AI News
Tencent’s Hy4 AI Beats DeepSeek in Blind Test by 213 Finance Pros

Tencent has released Hy4, a new large language model that topped rival open-source models in a blind test run with financial professionals. The company announced the model on November 4, 2026.

Tencent said it plans to release the model weights and system cards in the coming days. The company said the release puts a focus on cybersecurity and biological safety.

Finance Experts Rate Hy4 First

The blind evaluation was run by the Tencent Hy team with Tencent Financial Technology. It involved 213 financial professionals, including 153 experts from outside institutions. About 70 percent had more than five years of industry experience.

The testers judged the models across 579 real-world financial tasks. Hy4 scored 3.04, ahead of DeepSeek-V4.1-Flash-1017 at 2.97. Kimi K3.1 Pro and GLM-5.3 each scored 2.68.

Tencent said the model was trained on thousands of specialized workspaces and tens of thousands of tasks from real scenarios. These cover fund research, securities analysis and quantitative investment strategies.

Financial AI company Rabyte also tested the model on its internal Investment Research Agent Benchmark. Hy4 scored 12.03 percentage points higher than the earlier Hy4 Preview.

Tencent added search tools that focus on recent information and checking facts across sources. In internal tests, the hallucination rate fell from 12.95% in Hy4 Preview to 7.2%. Errors from unsupported assumptions dropped from 23.1% to 5.6%.

Hy4 is the first Tencent Hy model with visual understanding. It can process detailed images, analyze complex charts and solve math problems shown visually.


The model can also use computers for the first time. It works with both command-line tools and graphical interfaces. It received special training on professional software such as Blender and CAD.

On Tencent’s WorkBuddy platform, Hy4 posted an average score of 83.18 and a pass rate of 96.15 percent. Tencent said it ranked first in nine of twelve scenarios in the mainland Chinese market.

In game development, Hy4 works with Unreal Engine, Unity and Godot. It improved 19.1 percent on GameDev benchmarks and 9.8 percent on GameLogic benchmarks compared with Hy4 Preview.

Tencent also open-sourced two research models built with Hy4 and its Hyra 2.0 harness. HyCell handles single-cell representations, and HyPDE covers physics fields.

Security experts from Tencent’s Xuanwu Lab and Keen Security Lab tested the model. They found no major concerns about its offensive cyber abilities in the areas reviewed. Tencent said offensive exploitation material was left out of the training data.

Pricing Stays Flat as Free Access Opens

Hy4 keeps the same pricing as Hy4 Preview. It costs about US$0.90 per million input tokens and US$2.70 per million output tokens. Cached input tokens cost about US$0.045 per million.

The model is available on Tencent Cloud TokenHub, WorkBuddy, CodeBuddy, Yuanbao and other Tencent products. Its API is also offered through third-party platforms, including OpenRouter, Hermes and OpenCode.

Tencent is working to add the model to QQ, Honor of Kings, Game of Peace, WeRead and Tencent News. A lower-cost version called Hy4 Flash is set to launch soon.

WorkBuddy and CodeBuddy are free to use from November 4 to 15. Hy4 is also 50% off on Tencent Cloud TokenHub through December 31, 2026.

From our research desk
AI Jobs Automation Index
Which jobs are AI tools targeting most? We mapped 3,400+ AI tools to real job functions — with BLS employment & salary data.
Explore the index
maisiekooc

Written by

Maisie is a news writer at Agent Locker, covering the latest developments in artificial intelligence, emerging technology and the companies shaping the future.

Discover AI Agents