Tencent Hy4 goes GA on Novita AI at the same price as Preview
Novita AI has made Tencent Hy4 its official GA model under the ID tencent/hy4, keeping the price identical to Hy4 Preview while adding image input to the catalog entry. Tencent describes stronger agentic performance than the Preview release, a claim Novita advises customers to test on their own tasks.
- Tencent Hy4 is now the official GA model on Novita AI under the ID tencent/hy4.
- Pricing is unchanged from Hy4 Preview: $0.834 per 1M input, $2.501 per 1M output, $0.042 per 1M cached input reads.
- The GA catalog entry accepts text and images, while Preview accepts text only.
- Tencent claims stronger agentic and long-horizon performance, which Novita advises testing on your own tasks.
- Hy4 Preview stays listed separately with the same price and limits.
Pricing and limits unchanged from Preview
The GA model is served through Novita AI's OpenAI-compatible serverless API. According to Novita AI, it costs $0.834 per 1M input tokens, $2.501 per 1M output tokens and $0.042 per 1M cached input reads. These rates match the Hy4 Preview listing exactly.
Context and output limits also match on paper. The catalog lists a 1,000,000-token context window and a 64,000-token maximum output, while the model page displays 977K and 63K. Novita recommends using the model page figures when planning request budgets. The listing was checked on November 4, 2026.
What is different: image input and Tencent's capability claims
The main change is capability rather than cost. The catalog entry for tencent/hy4 lists text and image as input, while tencent/hy4-preview lists text only. Output remains text.
Tencent's release description says the official Hy4 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. Novita stresses that this is the vendor's own evaluation, not a guarantee for any particular workload.
Novita's model detail page does not document a separate vision flag or image-specific limits. Images are passed through the standard chat completions content-part format, as a URL or base64 data in an image_url part. The company suggests sending one representative screenshot first to check quality and token cost.
API access and features
Hy4 runs as a serverless model with FP8 quantization. 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. Existing OpenAI or Anthropic-style clients can therefore address it without a new transport.
Listed features are function calling, structured output, reasoning and serverless delivery.
Both IDs stay live: what it means for users
Hy4 Preview remains listed separately with the same pricing and limits. Requests to tencent/hy4 do not fall through to tencent/hy4-preview, and the reverse is also true, so Novita advises naming the ID in configuration and logs.
Because price and limits are the same, Novita describes migration as mostly a configuration change. It suggests switching the model field in one environment first, keeping the Preview ID as a fallback, and replaying a fixed evaluation set of the hardest agentic and long-context cases. Metrics to compare include task success, tool-call validity, retries, latency and total token cost.
Novita sees the model as a candidate for coding agents, long-horizon workflows, long-context analysis and image-assisted developer tasks. It points to short, high-volume text jobs, text-only workloads without agentic demands and latency-sensitive features as cases where a cheaper model or Preview may serve as well. It also notes that GA is an official release, not a long production history. Since the price did not change, the decision to move rests on measured quality.
