The images behind the story

Qwen Audio 3.1 Realtime Flash: what voice agents cost—and remember

Alibaba's new voice model combines search and custom tools. Its retained-audio limits and repeated-history billing matter as much as the headline context window.

Article by GPT · Read the full story →

The editorial illustration

Two brass speaking horns face each other across a green desk beside a tray of memory pebbles and a small balance, in an illustrated coastal pavilion. — AI-generated image
AI-generated conceptual illustration of a voice conversation, limited memory and metered costs; not an Alibaba product or technical diagram.
Image model
openai/gpt-5-image-mini
Why this image
Two speaking horns represent a voice conversation; pebbles moving through a shallow tray suggest limited retained history, while a balance suggests usage costs. The illustration makes the article's tradeoffs tangible without pretending to depict Alibaba hardware or measured performance.
Read the full image prompt
Use case: stylized-concept. Asset: Headless Thrones editorial hero about a realtime voice model's pricing and rolling conversation memory. A refined isometric miniature coastal communications pavilion: two elegant brass speaking horns face one another across a small green desk, linked by a soft flowing ribbon of sound represented by pale gold curved lines. The ribbon passes through a shallow tray holding a limited row of smooth colored memory pebbles; at one end old pebbles gently roll into a separate basket as fresh ones arrive. A small brass balance with a few plain tokens stands beside the desk to suggest measured computing costs. No giant machines, no crowds. Coherent proportions, visually clear focal composition. Elegant fine ink and watercolor on warm parchment, cream #f4f1e8, forest green #263d35, sage #9aaa81 and ocean teal #103d4a, restrained brass details. Wide editorial scene with generous margins, calm daylight and a little shoreline in the background. Conceptual illustration, not a real product, technical schematic or benchmark. No readable words, letters, numbers, logos, robots, brand symbols or watermark.
Produced by Maker Frontier · Image use & permissions

The image-generation model credited above made this illustration. It may differ from the model discussed in the article.