Tencent Hy4 is live on Together AI
- Source
- together.ai
- Date

Context
Together AI, a company that hosts AI models and sells access to them through its API, now serves Tencent's Hy4, according to Together's launch post of November 19, 2026. Hy4 is the flagship of Tencent's Hy model family and the first Hy model to accept images alongside text. On Together it runs on serverless endpoints, where customers pay per , or on dedicated endpoints with reserved capacity. It has a 1M-token and supports and JSON output, at $0.83 per million input tokens, $2.50 per million output tokens and $0.042 per million cached input tokens.
The post's evidence for what the model can do is two tests Together ran itself, not a benchmark. Given a screenshot of a bar chart from an earlier Together post, Hy4 returned the chart's title and all three values as JSON, then rebuilt the chart as a single HTML file that Together says matched the layout, labels, values, colors, dashed ceiling line and footnote. From a short text prompt, it wrote a browser game that Together opened without changing the code, played to the game-over screen and restarted. These show the image input and code generation working on one task each. They say nothing about how often the model gets such tasks right.
Independent measurement already exists. Earlier Intel covered Artificial Analysis' benchmark run, which scored Hy4 35 on its Intelligence Index, level with Grok 4.7 and 1 point behind Claude Haiku 5.5, and found it strong on agentic coding. The same run measured an 89% rate on its knowledge test, against 37% for Kimi K3.1 Pro. That test is not chart reading, but it is a reason to check figures Hy4 extracts rather than pass them straight into a report.
Together is also not the only place to run the model: Vercel's AI Gateway added Hy4 earlier in November. Tencent's announcement describes more than Together demonstrates, including computer use in command-line and graphical interfaces, and said Tencent planned to release the weights and system cards in the following days. Together's post does not say whether computer use works on its endpoints. What this post establishes is narrower: Hy4 can now be called on Together at the listed price, and the case for its quality rests on outside benchmarks, not on these two demonstrations.
- In Together AI's own test, Hy4 read a screenshot of a bar chart and returned the title and all three bar values as JSON exactly as they appear in the chart.
- Asked to rebuild that chart as a single HTML file, Hy4 reproduced the layout, labels, values, colors, dashed ceiling line and footnote; only the typeface differed, because Together requested generic fonts.
- From one text prompt, Hy4 wrote a browser game that Together ran and restarted without editing the code, and it added a title screen and a best score kept across rounds that the prompt never asked for.
- Hy4 is the first model in Tencent's Hy family to accept images. On Together it pairs a 1M-token context with tool calling and JSON output at $0.83 input, $2.50 output and $0.042 cached input per million tokens.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
- hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
Checking sign-in…
Loading comments…


