Tencent Hy4 goes live on Together AI with 1M-token context and image input

Together AI has made Tencent's flagship Hy4 model available through serverless and dedicated endpoints. Hy4 is the first model in Tencent's Hy family to accept images alongside text, and it offers a 1M-token context window.
- Tencent's Hy4 is live on Together AI via serverless and dedicated endpoints under the model string tencent/Hy4.
- It has a 1M-token context window, text and image input, tool calling and JSON output.
- Pricing is $0.83 per 1M input tokens, $2.50 per 1M output tokens and $0.042 per 1M cached input tokens.
- Hy4 is the first model in Tencent's Hy family to accept images alongside text.
- In Together's tests, Hy4 extracted chart data as JSON, rebuilt the chart in HTML and wrote a playable browser game.
Pricing and capabilities
The model is called with the string tencent/Hy4 through Together's chat completions API. According to Together AI, it takes text and image input and returns text, and it supports tool calling and JSON output.
- Input: $0.83 per 1M tokens
- Output: $2.50 per 1M tokens
- Cached input: $0.042 per 1M tokens
Developers can start on serverless infrastructure and move to a dedicated endpoint when they need reserved capacity. JSON output is enabled by setting response_format to {"type": "json_object"}, and tool calling works by passing functions in the standard tools parameter, Together says.
Two tests: a chart and a game
Together AI ran two small tests to show what Hy4 does in practice. The first targeted vision, the capability Hy4 adds to the Hy family. The team gave the model a screenshot of a chart from its Together Storage Accelerator post and asked for the contents as JSON. According to Together, the response returned the title and all three values exactly as they appear in the chart: a TSA warm read at 3.18, an R2 origin direct read at 0.16 and a TSA cold read at 0.14.
The team then asked Hy4 to reproduce the same chart as a single HTML file, preserving the visual layout as well as the numbers. Together says the model reproduced the layout, labels, values, colors, the dashed ceiling line and the footnote. The typeface differs because the prompt asked for generic fonts.
The second test used a text prompt only. Hy4 was asked to write a small, original browser game in one HTML file, with no libraries or external assets, keyboard controls, a score, a way to lose and a restart. It returned a game called Polarity. Blocks fall from the top of the screen, the player catches those matching their color and presses Space to switch colors, and a yellow hazard ends the run if it reaches the player.
Together reports that it opened the file in a browser without changing the code. Hy4 also added a title screen explaining the rules and a best score that carries over between rounds, neither of which was requested. The team played a round, reached the game-over screen and restarted successfully.
How to try it
Together provides a Python example that sends the chart screenshot as a base64-encoded image_url block and requests a JSON object with a title and a list of bars. It requires installing the Together Python SDK, setting a TOGETHER_API_KEY and saving the chart as chart.png. Those who prefer not to write code can open the model in Together's playground, and the model page has the full details.
Together points readers to Tencent's own announcement for the wider release, including Hy4's new computer-use capabilities.
What it means for developers
For developers already building on Together AI, Hy4 adds a long-context model with image input and structured output behind one API. Together's published per-token prices, including a lower rate for cached input, make costs straightforward to estimate. The chart and game tests are Together's own demonstrations rather than independent benchmarks, so teams will want to check the model against their own workloads.
Together AI: Tencent Hy4 is live on Together AI

