AI Agenda #36: New Models and Investments

New Models and Protocols

This week in the AI ecosystem, new models were announced, especially focused on multi-agent systems and multimodal capabilities. The key developments are:

Nvidia Nemotron 3: A new open-source model family optimized for multi-agent systems was introduced.
OpenAI GPT Image 1.5 & GPT-5.2-Codex: While image generation capabilities for ChatGPT were updated, the software engineering-focused Codex version was released.
Google Gemini 3 Flash: Focused on low cost and high speed, this model was made the default in search and Gemini applications.
AI2 Molmo 2 & Olmo 3.1: Molmo 2, capable of video analysis (object tracking, counting), and the open reasoning model Olmo 3.1 were released.
Black Forest Labs FLUX.2: A high-end image model capable of generating visuals with real-time web data was announced.
Klarna Agentic Product Protocol: An open standard protocol enabling AI assistants to access more than 100 million products was introduced.
Alibaba Wan2.6: A multimodal model with storyboard and character reference support, capable of generating HD video up to 15 seconds.
xAI Grok Voice Agent API: A speech-to-speech voice technology API for developers was offered.
Meta SAM Audio: A model capable of separating specific sounds from audio files using visual and text inputs was released.

New Features and Integrations

Existing platforms and applications continue to integrate AI capabilities into workflows:

Google Translate & CC: While the translation service became context-focused with Gemini infrastructure, the email-based assistant CC began testing.
Zoom AI Companion 3.0: The new version connecting meeting data to automated workflows was introduced.
Adobe Firefly: Text commands and camera control features were added for video editing and assembly processes.
OpenAI 3rd-Party Applications: Support for third-party applications not requiring external installation was added inside ChatGPT.
YouTube Playables Builder: A Gemini 3-powered platform enabling game creation with text and visual prompts was announced.

Research Notes

Data from Stanford HAI and independent research reveals the competency levels of models and future expectations:

CFA Exam Success: It was stated that GPT-5, Gemini 3.0 Pro, and other leading models passed all three levels of the CFA certification exams. Gemini 3.0 Pro achieved a 97.6% success rate at Level 1.
2026 Forecasts: Stanford HAI predicts that AI use in healthcare will increase, but AGI (Artificial General Intelligence) expectations will not be realized in the short term.

Investment Agenda

Capital flow in the sector continues with major valuations and strategic partnerships:

Amazon & OpenAI: It was reported that Amazon is preparing to invest $10 billion in OpenAI and that the valuation could exceed $500 billion.
Databricks: It received more than $4 billion in investment at a $134 billion valuation.
Lovable: It reached a $6.6 billion valuation by raising $330 million in its Series B round.
Other Investments: Chai Discovery ($130M), Neural Concept ($100M), PolyAI ($86M), and Mirelo ($41M) completed investment rounds.