Tencent Hy4 is now the official GA model on Novita AI, listed under the model ID tencent/hy4 and served through the OpenAI-compatible serverless API. It costs exactly the same as Hy4 Preview: $0.834 per 1M input tokens, $2.501 per 1M output tokens, and $0.042 per 1M cached input reads, with the same 1,000,000-token context window and 64,000-token maximum output. The practical change is production status and capability. The catalog entry now accepts images alongside text, and Tencent describes stronger agentic performance than the Preview release.
Tencent Hy4 Specs and Pricing on Novita AI
| Field | Details |
|---|---|
| Display name | Tencent Hy4 |
| Model ID | tencent/hy4 |
| Base URL | https://api.novita.ai/openai |
| Endpoint family | chat/completions, responses, Anthropic-compatible |
| Input modalities | Text, image |
| Output modality | Text |
| Context window | 1,000,000 tokens listed; the model page displays 977K |
| Max output | 64,000 tokens listed; the model page displays 63K |
| Input price | $0.834 per 1M tokens |
| Cached input read | $0.042 per 1M tokens |
| Output price | $2.501 per 1M tokens |
| Release status | GA, serverless |
| Pricing vs Hy4 Preview | Identical input, output, cache-read, context, and max output |
Specs and pricing from the Novita AI model page and model catalog, checked November 4, 2026. The catalog lists nominal limits; use the model page’s 977K and 63K figures when planning request budgets.
Key Takeaways
tencent/hy4is the GA model on Novita AI. It is the production-facing release, whiletencent/hy4-previewremains listed separately.- The price did not move. Input, output, and cached input reads are unchanged at $0.834, $2.501, and $0.042 per 1M tokens.
- Limits match on paper. Both list a 1,000,000-token context and 64,000-token maximum output, with the model page showing 977K and 63K.
- Input support is wider. The GA catalog entry accepts text and images; the Preview entry accepts text only.
- The upgrade case is capability, not cost. Tencent claims stronger agentic results, so evaluate on your own tasks before moving production traffic.
What Changed from Hy4 Preview to Hy4 GA
Tencent’s release description says the official Hy4 release delivers substantial improvements over Hy4 Preview, with particularly strong gains in agentic capabilities and long-horizon productivity tasks, and introduces visual understanding for the first time. That is the vendor’s evaluation, not a guarantee for your workload. The useful question is narrower: does GA do better on the tasks you already run, and does it now handle image inputs you previously sent to a separate model?
Both IDs are live. Hy4 Preview is still listed with the same pricing and limits as GA. A request to tencent/hy4 does not fall through to tencent/hy4-preview, and the reverse is also true, so your configuration and logs should name the ID you use. The earlier Tencent Hy4 Preview launch guide remains the reference for Preview-specific, text-only behavior; use this page for GA.
Does Tencent Hy4 Support Image Input on Novita AI?
Check this against the endpoint listing rather than the release copy. The Novita AI catalog entry for tencent/hy4 lists input capabilities as text and image, while tencent/hy4-preview lists text only. That is a concrete difference between the listings, and the clearest signal that the visual-understanding claim reached the hosted configuration.
The model detail page does not document a separate vision flag or image-specific limits. Image content goes through the standard chat completions content-part format, with the image passed as a URL or base64 data in an image_url part. Before building an image workflow, send one representative screenshot through that format and confirm quality and token cost.
Tencent Hy4 API Access on Novita AI
Hy4 runs as a serverless model. The model page lists chat/completions and responses endpoints under https://api.novita.ai/openai, plus an Anthropic-compatible endpoint at https://api.novita.ai/anthropic, so existing OpenAI or Anthropic-style clients can address it without a new transport. The listed features are function calling, structured output, reasoning, and serverless delivery, with FP8 quantization.
For request fields and content-part syntax, use the current Create Chat Completion reference. If you want a request walkthrough for Tencent’s earlier Hy model, the Hy3 API quick start covers the chat completions shape; Hy4 has its own model ID, limits, and image support.
When to Use Tencent Hy4
- Coding agents and multi-step tool use. A reasonable candidate for repository-scale tasks, tool-call loops, and test-driven iteration where a weaker model loses the plan.
- Long-horizon productivity workflows. Ticket triage, document-heavy analysis, and multi-stage planning need the consistent context this model targets.
- Long-context analysis. The 1M-token window fits requests that combine specifications, repository state, and prior tool results, if you curate what goes in.
- Image-assisted developer tasks. Screenshots, architecture diagrams, and document pages can go in the same request as the instruction.
- Teams already on Hy4 Preview. If Preview passes your evaluation, GA keeps the same price and limits while adding official status and image input.
When Not to Use Tencent Hy4
- Short, high-volume text tasks. Classification, routing, and low-stakes rewrites rarely need a large MoE model; a cheaper model is the better default unless quality is the bottleneck.
- Text-only workloads with no agentic pressure. If Preview or a smaller model passes every test and images are not involved, GA offers no measurable reason to move.
- Hard latency floors. Measure time to first token, retries, and tool-cycle time before routing interactive features to a new release.
- Teams that cannot evaluate first. Tencent’s claims are vendor benchmarks; without your own task set and acceptance criteria, a switch is a guess.
- Workflows that need a long production record. GA is an official release, not a long-running production history.
How to Move from Hy4 Preview to Tencent Hy4
Migration is mostly a configuration change because pricing and limits match. Move the model field from tencent/hy4-preview to tencent/hy4 in one environment first, keep the Preview ID as a fallback, and replay a fixed evaluation set covering your hardest agentic and long-context cases. Compare task success, tool-call validity, retries, latency, and total token cost.
There is no cost-model change to justify the switch, so the decision rests on measured output quality. If you send images, test that path separately, and keep the model ID in configuration so you can roll back without a deploy.
Conclusion
Tencent Hy4 on Novita AI is the GA version of a model whose price, context, and output ceiling already matched Preview. You are not buying a cheaper or larger model; you are testing whether the official release does better on agentic, long-horizon, and image-assisted work. Start with your own evaluation set, keep Preview as the fallback, and move traffic only where the measured gain justifies it.
FAQ
What is the model ID for Tencent Hy4 on Novita AI?
The exact model ID is tencent/hy4. Use it in the model field with the OpenAI-compatible base URL, and confirm the current listing on the Tencent Hy4 model page.
Is Tencent Hy4 the same as Hy4 Preview?
No. They are two separate model IDs that are both live on Novita AI. Hy4 Preview is the earlier preview release with text-only input; Hy4 is the official GA release, with the same price and limits but broader input support.
How much does Tencent Hy4 cost on Novita AI?
As of November 4, 2026, Novita AI lists $0.834 per 1M input tokens, $2.501 per 1M output tokens, and $0.042 per 1M cached input reads. These rates are identical to the Hy4 Preview listing; confirm them on the Tencent Hy4 pricing page before production budgeting.
What are the context window and maximum output for Tencent Hy4?
The Novita AI catalog lists a 1,000,000-token context window and a 64,000-token maximum output, while the model page displays 977K context and 63K output. Plan against the lower figures and verify the current model listing if your request sizes sit near the limit.
Does Tencent Hy4 support image input on Novita AI?
Yes. The Novita AI catalog entry for Tencent Hy4 lists input capabilities as text and image, unlike the text-only Hy4 Preview entry. The model page does not document a separate vision parameter, so use the standard chat completions image content part and test a representative image before relying on it.
Is Tencent Hy4 a good fit for coding agents?
It is a reasonable candidate. Tencent describes GA as stronger on agentic and long-horizon tasks than Preview, and the model supports function calling and structured output. Test it on your own failing tickets and multi-file changes before making it a production default.
