Tencent Hy3 Dropped Today. Here is What 6 Cents Per Million Tokens Actually Buys You.

2026-07-09 · Source: Towards AI - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

Tencent has released Hy3, a production-ready, 295-billion-parameter open-weights model, with only 21 billion active during inference. This model features a 3.8-billion-parameter Multi-Token Prediction head for speculative decoding, a 256K context window, and utilizes 192 experts with top-8 routing. Hy3, licensed under Apache 2.0, is immediately available on OpenRouter for approximately six cents per million input tokens, with a two-week free trial where the provider absorbs onboarding costs. It has reportedly outperformed GLM-5.1 in blind expert reviews of real work tasks, marking a significant entry into the market beyond a research preview. Its combination of price, performance, and license is notable.

Key takeaway

For Directors of AI/ML evaluating new model procurements, Tencent Hy3's combination of competitive performance, Apache 2.0 license, and six-cent per million token pricing demands immediate attention. You should explore its two-week free trial on OpenRouter to assess its fit for your team's specific tasks, especially for local or cost-sensitive deployments. This release could significantly alter your model strategy.

Key insights

A high-performance, cost-effective, open-weights LLM is now production-ready, disrupting existing procurement strategies.

Principles

• Open-weights models can achieve competitive performance.
• Cost-performance ratio is a key market differentiator.
• Production releases signal market maturity.

In practice

• Evaluate Hy3 on OpenRouter for two weeks free.
• Consider Hy3 for local, firewall-protected deployments.

Topics

• Tencent Hy3
• Large Language Models
• Open-weights Models
• Model Inference Costs
• Apache 2.0 License
• OpenRouter

Best for: CTO, VP of Engineering/Data, NLP Engineer, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Tencent's Apache-licensed Hy3 takes on GLM-5.2 at half the size — and wins everywhere except coding — Tencent's Hy3 offers a reliable, deployable open-weight MoE model with a permissive license, balancing performance with resource efficiency.
• Chinese AI models regularly pass 30 percent on OpenRouter as cost gap widens — Chinese AI models are rapidly gaining US market share due to their 60-90% lower cost, despite a 6-9 month capability gap.
• not much happened today — AI development is shifting towards multi-cloud distribution, cost-aware agentic workflows, and specialized hardware for inference.
• The Pulse: a new trend, smart model routing — Intelligent model routing is emerging as a critical solution to optimize LLM costs by dynamically selecting the most appropriate model for each task.
• tencent/Hy3 — Tencent's Hy3 MoE model delivers competitive performance against much larger models while maintaining a more efficient active parameter count.
• Tencent uses product rollout, not just benchmarks, to define Hy3 preview — Tencent's Hy3 preview emphasizes real-world product integration and performance over isolated benchmark claims.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards AI - Medium.