The ScrollThe Scroll
Back to The Scroll

The Model Wars Heat Up

Google's Gemini 2.0 arrives with agentic ambitions, OpenAI drops o1 for the masses, and the real story is what happens when these models hit production.

What Is Going On

Google bets on agents. Google launched Gemini 2.0 Flash with native tool use and 'agentic' capabilities. The interesting part isn't the benchmark numbers—it's Google betting that multimodal agents will matter more than raw text performance.

OpenAI's pricing signal. OpenAI released o1 to all ChatGPT users and dropped o1 Pro for $200/month. The pricing signals where they see value: not in the base model, but in extended reasoning chains.

Claude gets hands. Anthropic's Claude now handles computer use in beta. Early reports suggest it works better than expected on simple workflows, worse than expected on anything involving judgment.

AI Corner

Fine-tuning gets cheaper. Fine-tuning costs dropped 60% across major providers this quarter. This matters more than new model releases for most enterprise deployments.

RAG hits walls. Retrieval-augmented generation (RAG) is hitting scaling walls. Teams are discovering that chunking strategies matter more than embedding models—a lesson that feels obvious in hindsight.

News You Can Use

  • Bedrock guardrails. AWS announced Bedrock guardrails for agentic workflows. Essential for enterprises deploying autonomous systems.

  • Hugging Face goes lightweight. Hugging Face released a lightweight inference server that actually runs on reasonable hardware.

  • EU deadline looms. The EU AI Act's first compliance deadline is approaching. Most companies are not ready.

Get The Scroll in your inbox

If this kind of signal is useful, you can get it twice a week by email.

Subscribe