Hy4 Preview API: Pricing, Reasoning Tokens and Setup

Hy4 Preview is Tencent Hunyuan's current preview flagship, live as model ID hy4-preview. The published preview API rate is $0.834 per 1M input tokens and $2.501 per 1M output tokens.

Two operational facts matter more than the sticker. It is a preview — Tencent can change or withdraw it. And it spends heavily on reasoning tokens, so a too-small max_tokens can return an empty reply.

Hy4 Preview API pricing #

A $5 prepaid balance buys roughly 6.0 million input tokens, or 2.0 million output tokens, if a workload used only one category. Reasoning tokens are output. A short visible answer can still be an expensive one.

Hy4 vs Hy3 vs nearby flagships #

Hy3 on RouterPlex is not a vendor list price. Tencent publishes Hy3 as open weights without a first-party hosted API tariff; this catalog uses a public-market reference. Hy4 Preview is a published preview API rate.

Cost per agent turn #

One 40,000-token prompt returning 2,000 tokens, no cache, ignoring extra reasoning:

The arithmetic for Hy4: 40,000 × $0.834 / 1M + 2,000 × $2.501 / 1M = $0.03336 + $0.005002 = $0.03836. Extra reasoning tokens add $2.501 per 1M on top of that.

Set max_tokens high enough #

Hy4 Preview is a reasoning model. Reasoning tokens count against the output cap. If you leave max_tokens at a chat-sized default, the model can spend the whole budget thinking and return an empty content string.

Treat that as a configuration bug, not a model outage. Raise max_tokens, then check usage.completion_tokens_details.reasoning_tokens before you decide the route is expensive.

Hy4 Preview specifications #

Call Hy4 Preview with RouterPlex #

curl https://api.routerplex.com/v1/chat/completions \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hy4-preview",
    "max_tokens": 16000,
    "messages": [
      {"role": "user", "content": "Name the riskiest assumption in this migration plan."}
    ]
  }'

import os
from openai import OpenAI
 
client = OpenAI(
    api_key=os.environ["ROUTERPLEX_API_KEY"],
    base_url="https://api.routerplex.com/v1",
)
 
response = client.chat.completions.create(
    model="hy4-preview",
    max_tokens=16000,
    messages=[{"role": "user", "content": "Design a rollback-safe deployment plan."}],
)
 
print(response.choices[0].message.content)

Claude Code can point at the same model with ANTHROPIC_BASE_URL=https://api.routerplex.com and model hy4-preview. See the Claude Code setup guide.

Should you use a preview? #

• Evaluating Hunyuan 4 — yes, on a dedicated key with a hard budget and a high max_tokens.
• Production that cannot tolerate a withdrawn ID — wait for a generally available Hunyuan 4 route, or use GLM-5.3 / Qwen3.8-Max on the same key.
• Cheapest Hunyuan on this catalog — hy3 at the public-market $0.20 / $0.80, 256K context, no first-party tariff.

The live Hy4 Preview page is the number this route will actually bill.