ThursdAI - The top AI news from the past week · From Weights & Biases, Join AI Evangelist Alex Volkov and a panel of experts to cover everything important that happened in the world of AI from the past week

📆 ThursdAI - Dec 18 - Gemini 3 Flash, Grok Voice, ChatGPT Appstore, Image 1.5 & GPT 5.2 Codex, Meta Sam Audio & more AI news

·39 min·2 clips

Gemini 3 Flash brings pro-level intelligence at flash-level speed for 50 cents per 1 million input tokens.

1. ThursdAI’s December 18, 2025 episode covers the week’s AI releases, including Gemini 3 Flash, ChatGPT Images 1.5, Grok Voice, and GPT 5.2 Codex. 2. Alex Volkov hosts the show as an AI evangelist with Weights & Biases, and the guest lineup includes Wolfram Ravenwolf, Jan Pelik, MLGJ, Ryan Carson, Quindla Haltmann-Kramer, and Niston. 3. The episode exists to compress a three-and-a-half-hour live stream into a shorter weekly news recap for listeners tracking AI releases. 4. NVIDIA Nemotron 3 Nano is presented as a 30 billion parameter model with 3 active Mamba MoE layers, 25 trillion training tokens, and one million context length. 5. The hosts note that Nemotron 3 Nano reaches 99% on AIME25 with tools and that NVIDIA released pretraining and post-training datasets publicly. 6. Gemini 3 Flash gets framed as Google’s frontier-class flash model, priced at 50 cents per 1 million input tokens and built for low-latency use. 7. The discussion says Gemini 3 Flash now powers Google Search, Gemini Assistant, Gmail, Docs, Chrome, Android, and Google Home devices. 8. The hosts compare Gemini 3 Flash with Gemini 2.5 Flash and Gemini 3 Pro, emphasizing faster tool calling and stronger multimodal behavior. 9. They also say Gemini 3 Flash is available immediately across the Google stack, unlike older rollouts that arrived later or in fewer products. 10. OpenAI’s GPT Image 1.5 is described as four times faster and 20 times cheaper than the prior image model. 11. The team contrasts GPT Image 1.5 with Nano Banana, saying OpenAI’s model can be more inconsistent while sometimes taking more creative liberties. 12. Grok Voice is introduced as an API product with claims of 92% accuracy on Big Bench Audio and support for 100-plus languages. 13. The hosts mention Tesla integration, including steering-wheel voice activation and deep tools integration for navigation and other actions. 14. Chatterbox Turbo is presented as a 350 million parameter audio model with strong benchmark results, sub-150 millisecond latency, and zero-shot voice cloning. 15. Meta’s SAM Audio is described as a model for isolating sounds in video, such as separating a train noise from a person speaking or isolating drums from music. 16. The episode also covers Function Gemma, a 270 million parameter Google model designed for function calling on phones, browsers, and other small devices. 17. The tone is fast, collaborative, and benchmark-driven, with multiple hosts jumping in to compare models, prices, and product rollouts. 18. The format alternates between news rundown, live reactions, and short technical explanations rather than a single interview arc. 19. Anyone tracking weekly AI model releases and product integration will get the most value here. 20. Casual listeners who want a single deep-dive story may skip this week’s rapid-fire recap.

As heard by us

A brisk AI release roundup led by Gemini 3 Flash, with GPT 5.2 Codex arriving as late breaking news.

ThursdAI's December 18 update tracks a crowded week of AI releases that barely pauses for the holidays. Gemini 3 Flash gets the clearest strategic read, with the show tying its rollout across Google Home appliances and Android phones to Google's distribution advantage before…

Read the full review in PlayNext →

Why you'd press play

Want one week's AI whiplash: Gemini 3 Flash, ChatGPT Images 1.5, Nemotron Nano, Mistral OCR, and a GPT 5.2 Codex closer?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from ThursdAI - The top AI news from the past week