
Tool Retrieval in the Gemini API
Gemini's Interactions API now supports on-demand loading of tool parameter schemas. Agents with large tool libraries can keep schemas out of context until needed, cutting input tokens, latency, and selection errors.
Loading launches from 4 labs
Topic
Generating or editing images and video

Gemini's Interactions API now supports on-demand loading of tool parameter schemas. Agents with large tool libraries can keep schemas out of context until needed, cutting input tokens, latency, and selection errors.

Google's new Gemini 4 model pairs a 1M-token context window and four thinking levels with lower cost per task. It scores 61 on the AA Intelligence Index and targets complex coding, analysis, and cyber workflows.
A new community gallery lets users publish, explore, and remix custom interactive worlds created in Project Genie.
Google has updated the Genie 3 world model with 3-minute sessions, world chaining, 360-degree panorama grounding, and real-time promptable events.
Google rolled out Genie 4 to AI Ultra subscribers, featuring real-time natural language environment editing, native adaptive audio, and interactive world generation.

Dream Render is an AI web app that turns product descriptions or still photos into editable video drafts in minutes.

Google added Memory Summary to Personal Intelligence in Gemini. It gives users a single place to view, edit, or remove the saved preferences, interests, and ongoing goals Gemini uses to tailor responses.
Codex 0.168.0 lets you attach audio recordings when using audio-capable models. It skips confirmed rejected clips on later turns and uses clip duration to keep more short recordings in context.

Google's new Gemini 4 Hydrogen model features a 1M-token unified context window, four thinking levels, and long decode for agentic workflows.
Anthropic launched Claude Fable 5.2 and Mythos 5.2 with 1M token context, 128k max output, and always-on adaptive thinking. Fable 5.2 for reasoning agents; Mythos 5.2 for verified orgs. Pricing: $8/MTok in, $40/MTok out.
Meta rolls out real-time voice conversations and expressive avatars in the Muse app on mobile, web, and Mac, with hundreds of voice styles and real-time visual reactions. Available now in the US.

Google's new video generation model creates 15-second clips with multi-turn editing, 30% lower pricing, and better reference adherence, now generally available via API and Cloud.

OpenAI launched two new image models, GPT-Image-2.6 Sunburst and Flare, supporting adjustable complexity, SVG and animated GIF output, plus markup-guided editing.
Anthropic added message threads in beta to the Messages API, storing conversation history on the server. Subsequent requests only send new messages, reducing request size and time to first token.

Google's strongest frontier reasoning model for cybersecurity, long-horizon coding, and scientific research. It leads key vulnerability and coding benchmarks with a 1M-token unified context window.
Anthropic raised the maximum request size for the Messages API and Token Counting API from 32 MB to 128 MB on the first-party Claude API, allowing larger payloads in a single call.
Google introduced Genie 4, a new world model focused on realism in real-time simulation. Specifics on access, benchmarks, and capabilities were not detailed in the snippet.
OpenAI released GPT-6.1 Terra, a fast, efficient model built for visual reasoning, computer use, and natural writing. It is available through the OpenAI API.
OpenAI Codex 0.168.0 adds an opt-in feature flag to generate and edit animated GIFs with frame count and timing controls, expanding its image generation beyond static images.