腾讯 Hy4 上线 Together AI
Together AI 表示,腾讯 Hy4 现已通过其聊天补全 API 开放,支持无服务器和专用端点。平台列出该模型具备 100 万 token 上下文、文本与图像输入、工具调用及 JSON 输出。Together AI 还展示了图表提取、HTML 重建和浏览器游戏生成测试;这些属于平台自身的测试,而非独立基准测试。
Together AI 表示,此次上线让开发者可通过无服务器和专用基础设施调用 Hy4,其中专用端点可提供预留容量。平台还列出了长上下文、图像输入、工具调用和 JSON 输出等能力,但这些仍是厂商报告的主张。
TL;DR
- The model:
tencent/Hy4on Together AI, serverless and dedicated, with a 1M-token context window and text and image input. - The price: \$0.83 per 1M input tokens, \$2.50 per 1M output tokens, \$0.042 per 1M cached input tokens.
- Capabilities: image input, tool calling and JSON output.
Starting today, you can run Tencent Hy4 on Together AI through serverless and dedicated endpoints. Hy4 is Tencent's flagship model and the first in the Hy family to accept images alongside text.
We gave it two small jobs to see what it does in practice: extracting the data from one of our charts and rebuilding that chart in HTML, then writing a browser game from a single prompt. Here's what came back, along with the code to make the chart request yourself.
It read our chart, then rebuilt it
We started with vision, since that's the capability Hy4 adds. We took this chart from our Together Storage Accelerator post, gave Hy4 a screenshot of it, and asked for its contents as JSON.