Tencent's Hy3 AI Model Goes Global: Specs, Pricing, and What It Means for Enterprise AI

On August 5, 2026, Tencent announced a broad international rollout of its flagship Hy3 (formerly Hunyuan) large language model, making it accessible globally through multiple products, APIs, and cloud services . Here is a complete breakdown of what's available, how to access it, and why it matters.

How to Access Hy3 Globally

Tencent is distributing Hy3 through three primary tools :

• WorkBuddy — Tencent's AI-powered workplace agent, where Hy3 is integrated as the core reasoning engine. It is free for all global users through August 31, 2026 (Pacific Time) .
• Tencent Design Miora — the company's AI design studio platform .
• Tencent Cloud TokenHub — a cloud-based AI model marketplace for developers and enterprises .

Developers can also access Hy3 via API through OpenRouter (priced at approximately $0.1288 per million input tokens and $0.5336 per million output tokens) , and the model weights are published under the Apache 2.0 license on Hugging Face and ModelScope .

Key Specifications & Capabilities

Property	Value
Architecture	Mixture-of-Experts (MoE) — 192 experts, top-8 activated
Total parameters	295 billion
Active parameters per token	21 billion (+ 3.8B MTP speculative layer)
Context length	256K tokens
License	Apache 2.0 (commercially friendly)
Release date	July 6, 2026 (preview: April 23, 2026)

Hy3 is a hybrid fast-and-slow-thinking reasoning model designed for complex reasoning, instruction-following, coding, and AI-agent workflows. Tencent says it performs comparably to flagship models with 2–5× more parameters across reasoning and instruction benchmarks . The Hy3 preview debuted with hallucination rates at 12.5% and commonsense errors at 25.4%; the final July release dropped those figures to 5.4% and 12.7% respectively, after integrating feedback from over 50 internal Tencent teams .

Adoption Metrics: 68-Fold Growth and OpenRouter Dominance

Usage data reported by Tencent and confirmed by multiple sources shows remarkable early adoption :

• 68-fold API call volume increase over the previous-generation Hy2 model within its first week .
• Topped OpenRouter's global LLM usage leaderboard (#1 in tool-call volume, processing approximately 33 million autonomous agent executions daily, capturing an 8.9% share of global agentic traffic) .
• Average daily token consumption increased 20×, and WorkBuddy users selecting Hy3 preview grew 6× .
• The Hy3 preview held the #1 spot on OpenRouter's weekly rankings for three consecutive weeks starting April 27, 2026 .

Note: As of the August 5 announcement, Hy3 ranked third behind DeepSeek V4 Flash and Xiaomi MiMo-V2.5 on OpenRouter's weekly token usage ranking .

Connection to Tencent's Broader AI Strategy and the Yao Shunyu Hire

This global expansion is Tencent's first major overseas AI push since hiring former OpenAI researcher Yao Shunyu earlier in 2026 . Multiple reports frame the Hy3 international rollout as Tencent's most aggressive competitive move against Western AI leaders, signaling a strategic shift from a China-focused AI strategy to a global go-to-market posture . By combining open-source accessibility, enterprise tool integration, aggressive pricing (undercutting OpenAI and Anthropic), and a free-trial period on WorkBuddy, Tencent is aiming to capture developer and enterprise market share outside China — a direct response to the talent infusion and strategic direction set by Yao Shunyu's leadership .

Benchmark Performance

Hy3 scored 78% on SWE-bench Verified and 90.4% on GPQA Diamond . Tencent reports a 90% task resolution rate on internal agent benchmarks . These figures place Hy3 competitively among open-weight reasoning models, though independent verification remains limited.

Pricing Summary

Access Point	Pricing
WorkBuddy	Free until August 31, 2026 (Pacific Time)
OpenRouter API	~$0.1288 / M input tokens, ~$0.5336 / M output tokens
Official Tencent API	~1 RMB / M input, ~4 RMB / M output (about $0.15 / $0.59 USD)
Self-hosted	Minimum 8× GPU tensor parallel (H20-3e or H200-class)

Bottom Line for Developers and Enterprises

Tencent's Hy3 global rollout matters because it combines:

1. Open-weight accessibility under Apache 2.0, allowing custom fine-tuning and self-hosting
2. Cost-efficient pricing that undercuts many Western API providers
3. Enterprise product integrations (WorkBuddy, Miora, TokenHub) that reduce friction for adoption
4. Real-world traction evidenced by the 68-fold API surge and OpenRouter dominance

The model's primary strengths are in agentic workflows and coding tasks, areas where it has shown strong developer adoption. Its weaknesses include mid-tier performance on some general benchmarks and the need for high-end GPU clusters for self-hosting .