Hy4 Preview Lands on AI Gateway
💡Test a 770B open-source model with a 1M-token context window through one API.
⚡ 30-Second TL;DR
What Changed
Hy4 Preview has 770B total parameters and 49B active parameters per token.
Why It Matters
AI practitioners can evaluate a high-capacity open-source model through a unified API without building a separate provider integration. Its long context and coding focus may make it useful for agent workflows, although practical quality and infrastructure costs still need testing.
What To Do Next
Run a small coding-agent evaluation with tencent/hy4-preview in the AI Gateway playground, comparing context retention, latency, and cost against your current model.
Key Points
- •Hy4 Preview has 770B total parameters and 49B active parameters per token.
- •The model supports a 1M-token context window for long-horizon tasks.
- •Developers can call it with the AI SDK using the model identifier tencent/hy4-preview.
- •AI Gateway supports usage tracking, cost monitoring, retries, failover, routing rules, and Zero Data Retention.
🧠 Deep Insight
Background and context from public sources — not the original article. 10 sources cited.
🔑 Enhanced Key Takeaways
- •Hy4 Preview is served through Vercel's integration with third-party providers like OpenRouter, rather than being hosted directly on Vercel's own infrastructure.
- •The model has demonstrated superior performance in blind engineering benchmarks compared to domestic competitors GLM 5.3 and Kimi K3.
- •Tencent has explicitly positioned Hy4 for specialized scientific research, including molecular dynamics and condensed matter physics simulations.
- •The release is part of a broader Tencent strategy to integrate high-capability models into their proprietary productivity tools, specifically CodeBuddy and WorkBuddy.
- •The model is categorized as a 'frontier' entry in the Chinese AI market, directly challenging the market share of DeepSeek V4 Pro.
📊 Competitor Analysis▸ Show
| Feature | Hy4 Preview | GLM 5.3 | DeepSeek V4 Pro |
|---|---|---|---|
| Architecture | 770B MoE (49B active) | Dense/Hybrid | MoE |
| Context Window | 1M Tokens | 512K Tokens | 1M Tokens |
| Primary Focus | Scientific/Engineering | General Purpose | Coding/Reasoning |
🛠️ Technical Deep Dive
- Architecture: Mixture-of-Experts (MoE) design utilizing 770B total parameters with a sparse activation of 49B parameters per token.
- Context Window: Native support for 1M tokens, optimized for long-horizon software engineering and document analysis.
- Integration: Accessible via Vercel AI Gateway using the tencent/hy4-preview identifier, leveraging OpenRouter as the underlying provider layer.
- Optimization: Fine-tuned for high-complexity reasoning tasks including molecular dynamics and condensed matter physics.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (10)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Vercel News ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.