Video · Matthew Berman ·
Gemini Carbon, NVIDIA CEO AFFAIR, Fable 5.2, Meta Muse Ember, GPT Image 2.6, Minimax H3: AI NEWS
In short
The video surveys major AI model releases and industry developments, emphasizing Fable 5.2's clear capability lead over Gemini 4 Carbon despite Carbon's much lower cost. It also covers new open-weight coding and image models, faster generative tools, and the business and security implications of NVIDIA CEO Jensen Huang's reported affair investigation.
Key points
- Fable 5.2 ranks first on several benchmarks but is substantially more expensive and quota-intensive than Gemini 4 Carbon.
- Gemini 4 Carbon is a capable, inexpensive model that built an Excel clone in about 1 hour 40 minutes using roughly 26 million tokens.
- LongCat 2.5 has 1.6 trillion parameters, including 48 billion active parameters, but is too large for most local users.
- Ripple 65B Preview supports a 1-million-token context and scored 80.5% on Terminal-Bench 2.1 at $0.20 per million input tokens.
- Inkling 2 Small scores 31 on the Artificial Analysis Intelligence Index and competes with much larger open-weight models.
- Meta Muse Ember is a low-cost workhorse model, while Cognition's SWE-2 Flash targets fast delegated coding tasks.
- GPT Image 2.6 adds GIF generation, Seedream 5.5 offers highly realistic multi-turn image editing, and MiniMax H3 Image is an open-weight model available in an 18 GB quantized version.
Written by an AI model from the video’s captions and description.
In this story
A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.



