Video · WorldofAI ·
Gemini 4 Krypton BEATS FABLE 5.2!?, GPT-6.2 Preview, GLM-5.5 LEAKED & Claude Code UPDATE! AI NEWS
In short
The video surveys reported AI model releases, benchmarks, coding capabilities, agent features, and research breakthroughs across Google, Anthropic, OpenAI, DeepSeek, Qwen, Meta, and others. It matters because cheaper models and improved coding agents could change which systems developers use for everyday work, while many claims remain early or unconfirmed.
Key points
- Gemini 4 Krypton, reportedly tested as Gemini 4 Hydrogen, shows strong coding, simulation, SVG, and 3D-generation results near Claude Fable 5.2 in some tests.
- Claude Code adds custom themes, parallel project threads, `/by the way`, `/fork`, reminders, read-only sharing, and improved thread management.
- GPT-6.2 Astra previews reportedly show major gains in visual generation, web development, and agentic tasks.
- Google reports Gemini-assisted formalization of mathematical conjectures, including 1.8 million lines of Lean code across 29,000 sub-agent invocations.
- Anthropic recommends Opus 5.5 for daily coding, Fable 5.2 for difficult long-running work, Sonnet 5.5 for defined tasks, and Haiku 5.5 for lightweight operations; all support a 1-million-token context window.
- DeepSeek's API adds app-like web search, file handling, link processing, safety, voice, OCR, speech recognition, and text-to-speech capabilities, with paid subscriptions possibly forthcoming.
- Google Antigravity adds `/run-workflow` for reusable Python-orchestrated multi-agent workflows, while Qwen 4.0 Max Preview, Muse Spark 1.4, GLM 5.5, and Perplexity Teammates receive reported updates.
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.


