Tencent's open-weight MoE flagship — 770B total / 49B active parameters, a 1M-token context, and a productivity focus on coding, office work, and research.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hy4-preview",
"messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
}'Overview
Hy4 Preview is Tencent Hunyuan's open-weight flagship language model, released on August 28, 2026. It is a 770B-parameter Mixture-of-Experts with 49B parameters activated per token, a 1M-token context window, and training aimed at coding, office productivity, and scientific research. On Venice it runs with function calling, reasoning, and web search.
On Venice, Hy4 Preview runs under the private tier with zero retention — your prompts are not stored, profiled, or used for training, and there is no personal history tied to your usage. It is served quantized to fp4 to keep a 770B model this affordable per token. Note that Venice's privacy here comes from not keeping data, not from encrypted inference: TEE and end-to-end encryption are not part of this deployment.
Agent quickstart
The API is OpenAI-compatible: change the base URL and the model id and existing client code works unchanged.
curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hy4-preview",
"stream": true,
"messages": [{ "role": "user", "content": "Draft the release note." }]
}'curl https://api.venice.ai/api/v1/chat/completions \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hy4-preview",
"messages": [{ "role": "user", "content": "Find the rate limits." }],
"tools": [{
"type": "function",
"function": {
"name": "search_docs",
"description": "Search the API documentation.",
"parameters": {
"type": "object",
"properties": { "query": { "type": "string" } },
"required": ["query"]
}
}
}],
"tool_choice": "auto"
}'import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["VENICE_API_KEY"],
base_url="https://api.venice.ai/api/v1",
)
resp = client.chat.completions.create(
model="hy4-preview",
messages=[{"role": "user", "content": "Build without permission."}],
)
print(resp.choices[0].message.content)Specifications
Assessment
Use cases
Prompting
Enable the reasoning mode for hard coding or multi-step analysis tasks; switch it off for quick lookups to save tokens.
Define your tools once and let function calling drive multi-step agent loops — the model is tuned for it.
Request JSON schema output explicitly when you need machine-readable results instead of prose.
Paste whole files or modules rather than excerpts — with ~960K input tokens available, context is rarely the constraint.
Use web search prompts for anything time-sensitive instead of trusting the training cutoff.
Keep a stable system prompt and reuse prefixes so cached input bills at $0.05 per 1M tokens instead of $1.04.
Alternatives
| Model | Best for | Context | Open weights | Price (Venice) |
|---|---|---|---|---|
| Hy4 Preview | Cheap open-weight frontier for code and agents | 1M tokens | Yes | $1.04 in · $3.13 out / 1M |
| Kimi K3 | Open flagship with maximum headroom | 1M tokens | Yes | $3.75 in · $18.75 out / 1M |
| Qwen 3.8 Max | Long-horizon autonomous work | 1M tokens | No | $2.50 in · $7.50 out / 1M |
| Gemini 3.8 Flash | Fast, cheap closed assistant | 1M tokens | No | $0.94 in · $4.69 out / 1M |
Hy4 Preview is the right pick when you want open weights, a 1M-token context, and full tool use at the lowest output cost in its class — especially for high-volume coding and agentic pipelines. Pay more for Kimi K3 only if you need its extra headroom on the hardest reasoning tasks.
Pricing
Billed per token on Venice: $1.04 per 1M input tokens and $3.13 per 1M output tokens.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Getting a key
Create a key in API settings. Nothing else is required to start.
Export it as VENICE_API_KEY so the snippets above run unedited.
Point an existing OpenAI client at https://api.venice.ai/api/v1. The scheme is part of the value: an OpenAI client given a bare host does not resolve it.
Pass hy4-preview as the model and send the request.
FAQ
Hy4 Preview is Tencent Hunyuan's open-weight flagship language model, released August 28, 2026. It is a Mixture-of-Experts model with 770B total parameters and 49B activated per token, a context window exceeding 1M tokens, and training focused on coding, office productivity, and scientific research.
Venice bills Hy4 Preview per token: $1.04 per 1M input tokens, $3.13 per 1M output tokens, and just $0.05 per 1M cached input tokens. There is no subscription required — you pay only for what you use.
Yes. Tencent released Hy4 Preview under the Apache-2.0 licence, with weights available on Hugging Face and ModelScope. That means you can self-host or fine-tune it, or simply run it privately through Venice without managing any infrastructure.
Tencent released and open-sourced Hy4 Preview on August 28, 2026, and it has been available on Venice since August 2026. It is an early 'Preview' version of the Hy4 generation, so Tencent notes that further training improvements are still expected.
Yes. On Venice, Hy4 Preview supports function calling, structured JSON-schema output, a dedicated reasoning mode, and built-in web search. It is also code-optimized, which makes it well suited to agentic software-engineering workflows.
Both are open-weight models with roughly 1M-token contexts. Hy4 Preview costs far less per output token ($3.13 vs $18.75 per 1M on Venice), making it the better choice for high-volume coding and agent work. Kimi K3 is the pick when you need maximum capability on the hardest tasks and budget is secondary.
Venice runs Hy4 Preview under its private tier with zero retention: prompts and outputs are not stored, profiled, or used for training. You can use it in the Venice app or through the Venice API without building a personal usage history. Note that inference is not TEE-encrypted — privacy comes from Venice not keeping your data.
Hy4 Preview offers a 1,024K-token (about 1M) context window with up to 64K output tokens on Venice. Tencent Cloud documentation lists roughly 960K as the practical maximum input, so very long documents or large codebases fit in a single request.
No. Hy4 Preview is a text-only model — it does not accept image inputs. If you need vision, use a multimodal model on Venice instead; Hy4 Preview's strengths are code, long-context text, tool use, and reasoning.
One key, free to start, no credit card.