The show of

Friday, 20 November 2026.

Listen to the audio edition

Five minutes or shorter.

2:01
Read the transcript

Emergent FM, for Friday the 20th of November.

About ten cents a second. That is what Google says Genie 4 costs to stream a live world, with stereo audio included. Genie 4 makes video you steer as it plays, with sound. You start a world from text, a reference image, or both. Then you steer it with the WASD keys, the arrow keys, or text typed mid-session. Input costs a dollar fifty per million tokens. Text output costs nine dollars, and video output costs seventeen dollars fifty. It is on the paid tier of the Interactions API. If you build real-time video apps on the Gemini API, call it with a paid-tier key to test live world generation.

Second. If your voice agent speaks more than one language, this one is yours. OpenAI says GPT-Live-1.1 says words more clearly across languages. It also keeps accents steadier and transcribes better. You can use it in the Live session setup you already have. To pause inference when nobody talks, set up standby on your WebRTC sessions. It resumes when someone speaks.

Third, from earlier this week. Three dollars in, fifteen dollars out. That is what a million tokens cost on Gemini 4 Neon, on the Gemini API paid tier. Google says it brings top-level intelligence and finishes tasks fast, across software engineering, knowledge work and science. It has four thinking levels, and high is the default. It takes a context window of a million tokens. To start, read What's new in Gemini 4 Neon.

Fourth. A model family you can download and run yourself. Tencent says it has opened the Hy4 flash family, in three versions. The weights are public, under Tencent's own licence. Flash reads text and images and writes text. It handles a context window of a million tokens.

That's the brief. The full stories are at emergent dot F M. This was read by an AI voice.

The show's 4 items

See all shows