VercelEngineering PostOfficial Source

Hy4 Preview now available on AI Gateway

Tencent’s Hy4 Preview is now callable through Vercel AI Gateway and selectable in coding agents, adding an open MoE option with a 1M-token context window.

Vercel · Aug 28, 2026

Open Source Open Markdown Open JSON

Source Summary

Vercel AI Gateway now serves **tencent/hy4-preview**, an open-source Mixture-of-Experts model with **770B total parameters**, 49B active per token, and a **1M-token context window**.

Practical Implication

Builders can select it through the AI SDK or connect Claude Code, Codex, Cursor, and other agents using Vercel’s coding-agent setup. Its stated focus makes it worth evaluating on long-running repository tasks and large-document work.

Agent-Ready Context

Vercel AI Gateway now serves **tencent/hy4-preview**, an open-source Mixture-of-Experts model with **770B total parameters**, 49B active per token, and a **1M-token context window**.

Builders can select it through the AI SDK or connect Claude Code, Codex, Cursor, and other agents using Vercel’s coding-agent setup. Its stated focus makes it worth evaluating on long-running repository tasks and large-document work.

This is a preview announcement, not evidence of coding quality, latency, cost, or dependable use of the full context window. Benchmark it on your own agent loop before changing defaults.

Connected Context · Feed7 Judgment

Hy4 adds another million-token coding-agent route, distinguished by its open-source MoE configuration rather than demonstrated performance. Against an already crowded long-context gateway catalog, the signal increases selection choice but also makes repository-level comparisons of sustained-context quality, latency, cost, and reliability more necessary.

Qwen 3.8 Flash now available on AI GatewayQwen 3.8 Flash provides another 1M-context gateway option, so Hy4’s open MoE architecture must be evaluated against an adjacent long-context route rather than selected by context size alone.GLM 5.3 now available on AI GatewayGLM 5.3 targets similarly long-horizon coding with a 1M-token window, reinforcing the need for matched repository tasks to distinguish behavioral and token-efficiency differences.Laguna S 2.1 is now available on AI GatewayLaguna S 2.1 supplies an open-weight alternative with a paid 1M-context tier and thinking controls, giving Hy4 a relevant open-model comparison with different operating choices.unslothai/unslothUnsloth shows a path from evaluating open models through a hosted gateway to serving them locally, but adds hardware, security, and operational tradeoffs absent from Hy4’s gateway announcement.

Context Map

modelcodingresearch#open-models#coding-agents#model-selection

Uncertainty

This is a preview announcement, not evidence of coding quality, latency, cost, or dependable use of the full context window. Benchmark it on your own agent loop before changing defaults.