Gemini API's Tool Retrieval preview loads full tool schemas on demand. Names and descriptions still go in upfront.
Keep your busiest tools loaded and measure input tokens on the same task.
#GeminiAPI#ToolRetrieval#AIAgents#TokenBudget
https://t.co/kETTgQUSMx
FastVideo open-sourced Dreamverse Harness, a film workspace with an editable timeline. The team says it runs fully locally on DGX Spark.
Try recutting a scene without rerendering every shot.
#Dreamverse#FastVideo#AIFilmmaking#VideoEditing
https://t.co/lmvc1aOg7Y
Kling Image 4.0 debuted at No. 6 for editing and No. 7 for text-to-image on Artificial Analysis.
My first test would be a background swap: does it leave the product label alone?
#KlingImage#AIPhotoEditing#TextToImage#ProductPhotography
https://t.co/6pk6dxWcI1
Kling Image 4.0 ranks #7 in Text to Image and #6 in Image Editing, behind only OpenAI and Microsoft AI models
It generates images up to 8K, and its 2K price is under a third of the price of every OpenAI model ranked above it. In Image Editing, it sits just behind GPT Image 2.5 Flare (max) at #5 and ranks in the top 5 for 5 of 7 editing actions.
It is also a big step up from Kling Image 3.0 Omni, which ranked #86 in Text to Image and #67 in Image Editing before this launch. Kling Image 4.0 gains more than 225 Elo on both boards and improves on Kling Image 3.0 Omni in every editing action and Text to Image capability.
Congratulations to @Kling_ai on the launch of Kling Image 4.0!
See below for our detailed analysis in Image Editing and Text to Image 🧵
Horizon Robotics' LoFT reconstructs 3D scenes and camera motion from RGB video. The demos are up; the code isn't out yet.
Watch the street-scene demo and check what happens around moving cars.
#LoFT#3DReconstruction#VideoGeometry#ComputerVision
https://t.co/jo6Jh4pv1r
OpenAI's internal agent results: 79% success monitoring training runs, 25% writing inference and serving code.
Track pass rates by task type before you hand over the keys.
#CodingAgents#AgentEvals#AIResearch#OpenAI
https://t.co/Upi8hoEB9k
Tencent released Hy4 flash Base and FP8 weights under its Hy License. It's natively multimodal, with a 1M context window.
Test it on your contact sheets before trusting it to shortlist shots.
#Hy4Flash#OpenWeights#ContactSheets#VisionAI
https://t.co/fzBcLuHIyv
Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base.
224B (12B active), 1M context, natively multimodal.
Fast, efficient and built for everyday productivity.
It's yours now. And stay tuned for Hy4 coming up next!
🤗: https://t.co/1LFBctqTpT
🐙: https://t.co/GRlDDobNcz
📄: https://t.co/t4FDcwAY0k
Released under the Tencent Hy License.
See full terms here: https://t.co/ZkyWWkAOBl
Recraft ran the same 12 prompts through Seedream 5.5, V4.1 Pro and GPT Image 2.6 Sunburst. Same brief, very different coffee packaging.
Try your hardest product brief in Compare models.
#Seedream55#Recraft#ProductMockups#AIImageGen
https://t.co/Rekq8iojKs
Genie 4 lets eligible Google AI Ultra users change the weather mid-scene with a prompt. The audio adapts as you explore.
Try it to scout the mood for a shoot before building a set.
#Genie4#WorldModels#VirtualScouting#Previsualization
https://t.co/4dzYFttAGW
Claude Code's experimental Code Mode has Claude write JavaScript that chains tool calls instead of reading each result one by one. Off by default.
Try it on changes repeated across many files.
#ClaudeCode#CodeMode#CodingAgents#AIDev
https://t.co/z5nLd9QJAT
Replit Missions: set a goal, budget and boundaries, and a swarm of agents plans, builds and tests. Replit says it built DataCat in two days within a $3,000 budget.
Cap the spend before you delegate.
#ReplitMissions#AIAgents#AgentSwarm#VibeCoding
https://t.co/1ZqaioaDH7
MAI Realtime 1, Microsoft's first speech-to-speech model, starts at $0.01 per audio minute and, Microsoft says, beats GPT Realtime 2.1.
Price your voice demo before you build.
#MAIRealtime#VoiceAI#SpeechToSpeech#Microsoft
https://t.co/pxraJ640RG
MAI Image 2.7 took the top spot on Artificial Analysis image editing, per Microsoft. It edits one object's color or material without disturbing the rest; ~$35/1,000 images.
Test it on product retouching.
#MAIImage#AIImageEditing#PhotoAI#Microsoft
https://t.co/tEr84d1B5s
Microsoft Research's open-source TokenEconomizer learns from past sessions in one repo, cutting later coding-agent cost 18.4-30.9% without changing agent or model.
Track cost per session first.
#CodingAgents#TokenEconomizer#AIAgents#MicrosoftResearch
https://t.co/FbLmTSMveV
Qwen4-Max is out: 2.4T-parameter backbone, 71B active per token, native vision and a 1M-token context window, per Qwen's own blog.
Test it on your image-heavy workflows.
#Qwen4Max#MultimodalAI#VisionAI#LLM
https://t.co/NduwLtFC90
Lovable's new Drafts: a full copy of your app, design to database, with its own chat and preview. Test ideas without touching the live app; History can roll back after you accept.
Safe sandbox for client tweaks.
#Lovable#VibeCoding#AIAppBuilder#NoCode
https://t.co/yMDZUv75s9
NotebookLM can now turn your sources into a shareable site. Type "make me a site" in Chat, describe it, and get an interactive dashboard or tracker.
Turn a research pile into a client-ready page.
#NotebookLM#GeminiAI#NoCodeAI#AICreators
https://t.co/Hv5QiVNl6t
You can now turn your sources into interactive, shareable websites 🤯
Just type "make me a site" in the Chat with a description of what you want to build (interactive dashboards, project trackers, etc). Once built, open it in a new tab or share it directly with others.
Available globally on web for ALL (18+) users. Share your examples below!
Anthropic's guide to always-on agents: they run indefinitely and decide how and when to act across many sessions. It covers the core primitives and how they fit together.
Read it before you build one. #AlwaysOnAgents#Anthropic#AgentDesign#ClaudeAI https://t.co/zUHB9PNFLd
Claude Code tip from Lydia Hallie: a Stop hook that exits with code 2 sends stderr back to Claude and triggers a retry. Exit 1 is non-blocking, so Claude just stops.
Use exit 2 for your test gates. #ClaudeCode#AIAgents#DevTools#Hooks https://t.co/89y1cTTEKA
You can use a Stop hook to have Claude try again when a check fails, but make sure it exits with code 2!
Exit 2 sends Claude back to retry with the hook's stderr, exit 1 is non-blocking so Claude just finishes even when the check failed
Apple's new Health AI Usage Study with OpenEvidence: up to 200,000 US participants, age 21+, iPhone 15 Pro or later on iOS 26. Answers carry linked peer-reviewed citations.
Citations are the new trust bar. #AppleHealth#HealthAI#OpenEvidence#AIResearch https://t.co/mu13z9cWNW
A viral 1.4MB pure-HTML 3D game: ~700KB is bundled Three.js, the rest is procedural code. No images, models or audio files. The breakdown has a prompt template.
Try procedural assets before stock packs. #ThreeJS#ProceduralGen#GenAI#CreativeCoding https://t.co/Yebh5O1Iup