SmarToken AI gateway logo for one smart access token to DeepSeek, Kimi, Qwen, GLM, Doubao, ERNIE and other Chinese frontier models
Menu
HomeFeaturesPricingDocsPlaygroundRoutingOpenRouterFAQAPItopicLoginGet API Key
APItopic
Model explainer7 min read/Updated 2026-05-25

Tencent Hunyuan Hy3 preview: agent rebuild, fast-slow thinking and real-world gaps

Hunyuan Hy3 preview as Tencent's first model answer after Shunyu Yao rebuilt the Hunyuan research system. It is a 295B-total-parameter MoE model with 21B active parameters, a 256K context window and a fast-slow thinking design aimed at agents. The hands-on tests are balanced: Hy3 preview shows clear ReAct-style planning and tool routing, but still struggles with data reliability and complete final deliverables.

Key takeaways

  1. 01This page positions Hy3 preview as Tencent Hunyuan's first major model after Shunyu Yao rebuilt the research and infrastructure system.
  2. 02Its main technical frame is a 295B-total-parameter MoE model with 21B active parameters, 256K context and fast-slow thinking for Agent workflows.
  3. 03The hands-on tests are mixed: Hy3 preview can plan and route tools, but long-chain deliverables still need stronger data reliability and completeness checks.
Tencent Hunyuan Hy3 preview: agent rebuild, fast-slow thinking and real-world gaps video guide. A short SmarToken video for Tencent Hunyuan Hy3 Preview: Agent Rebuild, Fast-Slow Thinking And Real-World Gaps, focused on model knowledge, evaluation angles and practical takeaways.

Hy3 preview is a rebuild signal

Hy3 preview is the first model answer after Shunyu Yao joined Tencent and rebuilt the Hunyuan pretraining, reinforcement-learning and infrastructure stack.

That context is important. The page is not only reviewing one model. It is asking whether Tencent's reorganized AI system can now move faster and aim at real-world Agent tasks. Hy3 preview started training at the end of January and was released less than three months later, as reported here. Read that speed as a preview-stage signal: fast enough to test the new route, not enough to prove final dominance.

  • Read Hy3 preview as the first step of a rebuilt Hunyuan system.
  • Separate infrastructure progress from final product maturity.
  • Watch the formal Hy3 release and Tencent product integrations.
Design pointMeaningValidation step
MoE295B total parameters with 21B active per token.Measure latency and quality under real Agent calls.
256K contextLarge window for documents and workflows.Run grounding tests with known answer locations.
Fast-slow thinkingBlend cheap fast steps with deeper reasoning.Route tasks by difficulty and compare cost.
Product rolloutHy3 appears across Tencent products and TokenHub.Check availability, pricing and integration docs.

The model is designed around Agent work

Hy3 preview is built for Agent scenarios, with broad capability, realistic evaluation and cost-performance as three design principles.

That aligns with Shunyu Yao's public research focus on language agents and ReAct-style loops. Agent work needs more than one strong skill. It combines reasoning, tool use, long context, code, dialogue and verification. Hy3 preview's architecture and evaluation direction make sense if Tencent wants Hunyuan to become a router for complex workflows rather than a pure chat endpoint.

  • Test reasoning and tool use together.
  • Avoid single-benchmark conclusions.
  • Measure task success and cost in the same harness.

The first hands-on test exposes preview-stage brittleness

In the market-data task, Hy3 preview struggled with data access, used fallback or simulated data and missed the requested full memo.

This is the most useful criticism in the page. Long-chain agents can look busy while still failing the final deliverable. If a model cannot obtain reliable data or does not produce the required written output, the task is not done. The page turns that into a clear evaluation rule: judge agents by completed artifacts, not by visible tool activity.

  • Require explicit evidence status for every dataset.
  • Reject simulated data unless the task allows it.
  • Check the final output against every prompt requirement.

The second test shows real Agent routing promise

In the SkillHub research task, Hy3 preview searches, deepens the investigation, checks authentication documents, studies MCP server logic and then writes a technical explanation.

This is where the preview looks more convincing. The model does not simply answer from memory. It routes through search, documentation and technical concepts before composing the final document. That is closer to a ReAct loop: reason, act, observe and write. The key question for developers is whether the loop stays stable on their own tools and documents.

  • Inspect the sequence of searches and tool calls.
  • Check whether the final document cites the right docs.
  • Run multiple tasks to detect dead loops or shallow searches.

Tencent's product surface will decide the next proof

Hy3 preview is landing across Tencent Cloud, Yuanbao, ima, CodeBuddy, WorkBuddy, QQ, Tencent Docs and agent frameworks such as OpenClaw or OpenCode.

That distribution is Tencent's advantage. A model can improve faster when it is tested against real users, documents, code workflows and business tools. But broad distribution also raises the quality bar. Hy3 preview will need reliable data access, clean final deliverables, stable pricing and strong product feedback loops before it becomes a default route for enterprise agents.

  • Track product-specific task success.
  • Refresh pricing and availability before publishing buying advice.
  • Compare Hy3 with adjacent Chinese agent models on real workflows.

Common mistakes to avoid

Mistake

Treating one article as a final ranking

Why it hurts

Model releases, pricing, quotas and benchmark positions can change quickly.

Better move

Use the analysis as a shortlist, then run current checks against your own workload.

Mistake

Choosing by brand instead of task

Why it hurts

A strong chat model may still be weak for long documents, coding agents, multimodal work or low-latency routes.

Better move

Define the job first, then compare models with prompts, files or media that match that job.

Mistake

Copying claims without a current verification check

Why it hurts

Benchmark numbers, context windows, API names and prices may be dated or provider-specific.

Better move

Confirm high-impact details against official docs, model cards or live provider pages.

Read it as a model briefing, not a setup guide

View model catalog ->

Use this page to understand the model family, the evaluation angle and the current conversation around it. Then choose one or two realistic prompts, documents or media tasks and test whether the model behaves well in your own workflow.

FAQ

These questions reflect recurring reader concerns around Chinese model knowledge, evaluation and fast-moving model releases.

What is the main point of Tencent Hunyuan Hy3 preview: agent rebuild, fast-slow thinking and real-world gaps?

Hunyuan Hy3 preview as Tencent's first model answer after Shunyu Yao rebuilt the Hunyuan research system. It is a 295B-total-parameter MoE model with 21B active parameters, a 256K context window and a fast-slow thinking design aimed at agents. The hands-on tests are balanced: Hy3 preview shows clear ReAct-style planning and tool routing, but still struggles with data reliability and complete final deliverables.

How should readers use the Chinese model context here?

Use it as market and product context, then verify technical claims, pricing, quotas and release details against official pages or your own tests before making a decision.

Why is there a short video with the page?

The video gives a fast visual summary of the model story, while the written page carries the caveats, comparisons and practical checks.

References and verification

SmarToken tracks public model releases, technical reports, product announcements and market signals to keep this catalog useful.

Technical claims need to be treated as dated unless they are confirmed by current official model cards, technical reports or provider announcements.

Pricing, quota, availability and benchmark details can change after the review date, so production decisions should use current vendor pages and direct workload tests.

DeepSeek-R1 official repository and technical report linksUsed for R1 release context, reinforcement-learning positioning and distillation caveats.Qwen3 official announcementUsed for Qwen3 model-family context, hybrid thinking and multilingual/app workflow claims.GLM-4.5 official announcementUsed for GLM-4.5 agent, reasoning and coding positioning.Hunyuan TurboS technical reportUsed for Hunyuan TurboS efficient reasoning, hybrid architecture and context-window claims.
Get API Key