🤖 AI Agents Weekly: Gemini 2.5 Flash Image, gpt-realtime, Anemoi Agent, Fine-tuning LLM Agents, Codex Updates, Agent Client Protocol

Gemini 2.5 Flash Image, gpt-realtime, Anemoi Agent, Fine-tuning LLM Agents, Codex Updates, Agent Client Protocol

In today’s issue:

• OpenAI introduces gpt-realtime

• Google launches Gemini 2.5 Flash Image

• Anthropic pilots Claude for Chrome

• OpenAI revamps Codex CLI

• Zed announces Agent Client Protocol

• Prime Intellect launches Environments Hub

• Fine-tuning LLM agents without fine-tuning LLMs

• xAI has launched grok-code-fast-1

• Microsoft announces MAI-Voice-1 and MAI-1-preview

• Top AI papers, research, and tool updates.

---

---

Top Stories

Gemini 2.5 Flash Image

Google introduced Gemini 2.5 Flash Image, its new state-of-the-art image generation and editing model, now available in preview via the Gemini API, Google AI Studio, Vertex AI, and OpenRouter. Building on Gemini 2.0 Flash’s speed and affordability, this release emphasizes higher quality outputs and fine-grained creative control.

• Character consistency – Maintains the same character or object across edits and prompts, enabling coherent storytelling, product showcases, and consistent brand assets.

• Prompt-based local editing – Supports natural language edits such as background blurring, object removal, pose changes, or colorization, demonstrated through customizable AI Studio template apps.

• World knowledge integration – Uses Gemini’s semantic understanding to handle tasks like interpreting diagrams, assisting with educational content, or making factually consistent edits.

• Multi-image fusion – Can merge and restyle multiple images in one prompt, useful for product mockups, scene composition, and design variations.

• Ecosystem integration – Launches with developer tooling in Google AI Studio (“build mode”), partnerships with OpenRouter and fal.ai, and mandatory SynthID watermarks for provenance.

Pricing is $30 per million output tokens (~$0.039 per image). The model is optimized for developers needing low-latency, affordable image generation while offering advanced controls for professional and creative use.

Blog