Tencent

2 models with a reviewed guide

All models

Tencent’s Hunyuan line on Y-API has two text-only entries: Hy3, a free 295B-parameter MoE with 21B active, and Hy4 preview, which scales to a 770B backbone with 49B active and a million-token window. Both target grounded, tool-driven work.

Sources reviewed: Y-API catalog snapshot:

About this vendor

Vendor-level selection advice by Y-API. The per-model pages linked below carry the model-specific checks; nothing here is a measured comparison.

Where this vendor fits

Hunyuan provides one of the catalog’s free text entries, and Hy4 preview is the cheapest million-window entry in its weight class — a combination that makes the family a low-risk pilot ground for grounded extraction work.

What to watch for

Hy4’s preview designation is a rollout caveat: catalog availability is not a stability promise. Hy3’s direct-answer default may or may not survive the gateway route, so reasoning settings need their own verification.

Models on Y-API

Prices are account-credit snapshots from the Y-API catalog. Each model links to its own guide with a cost estimator and a request example.

Tencent · Models on Y-API
ModelContext windowInput / 1M tokensOutput / 1M tokens
Hy3tencent/hy3262,144Free — no credit deducted
Hy4 previewtencent/hy4-preview1,048,576$1.00$0.10$3.00$0.30

Account credit / Cash equivalent · $1 paid adds $10 of credit; the cash equivalent is credit consumed ÷ 10. These are Y-API prices, not the publisher’s list prices.

Choosing within the lineup

Start extraction and multi-turn constraint work on free Hy3. Promote to Hy4 preview when tasks outgrow its 262,144-token window — behind a feature switch, keeping the tested Hy3 path until the pilot earns the migration.

Before you choose

Can production workloads run on Hy4 preview?

Possibly, but plan for churn: a preview entry can change behavior or availability without the notice a stable tier gets. Keep a fallback ID, a regression set and a way to halt a run, and re-run the comparison after each upstream update.

Sources & scope

Vendor scope on this site is exactly the reviewed model guides listed above; a vendor page exists only while at least one of its models is reviewed. Claims about architecture or positioning follow the cited pages. We do not publish measured latency, uptime or benchmark scores for any vendor.