Gemini Robotics ER 2 and Inkling-Small top releases, plus Vercel agent tools

What shipped

On 30 July 2026, new models and platform updates arrived across major hubs. Hugging Face and Replicate added several ready-to-run models while Vercel and Google shipped agent and robotics tools. Price reductions at OpenAI and fresh industry reporting round out the day.

Hugging Face trending

Four models climbed the Hugging Face trending list today. Thinking Machines and AMD each contributed one entry while EschaLabs and Audio8 added the others. The mix covers image-text, text generation, and speech tasks with direct fine-tuning paths on the hub.

• •Inkling-Small Thinking Machines released Inkling-Small, a compact image-text-to-text model that matches larger versions on reasoning tasks while using less compute for Vibe Builders testing visual agents.
• •Qwen3.6-35B-A3B-Escha-W2 EschaLabs released Qwen3.6-35B-A3B-Escha-W2, a text-generation model now available for direct download and fine-tuning on Hugging Face for developers needing open weights.
• •Audio8-TTS-Preview-0.6b Audio8 released Audio8-TTS-Preview-0.6b, a text-to-speech model that supports quick inference and fine-tuning for SMB owners adding voice output to apps.
• •Instella-MoE-16B-A3B-Think AMD released Instella-MoE-16B-A3B-Think, a text-generation model with mixture-of-experts architecture now live on the hub for production-scale text tasks.

Vendor launches

Vercel shipped the largest set of updates today across AI Gateway, Sandbox, and deployment speed. Google added Gemini Spark browsing and a new robotics model while Shopify and OpenAI contributed pricing and framework changes. The updates focus on running multiple agents and lowering inference costs.

• •Gemini Spark Chrome integration Google added web browsing to Gemini Spark so users can pull live page data directly into agent workflows without extra tools.
• •Inkling Small on AI Gateway Thinking Machines placed Inkling Small on Vercel AI Gateway where it delivers comparable results to the larger version at one-quarter size for lower-cost visual reasoning.
• •Multi-agent Sandbox support Vercel Sandbox now runs multiple isolated agents as separate Linux users so teams can test side-by-side agent systems without file conflicts.
• •GPT-5.6 price cuts on AI Gateway OpenAI lowered GPT-5.6 Luna input prices 80 percent and Terra 20 percent on Vercel AI Gateway with no markup passed to users.
• •Gemini Robotics ER 2 Google released Gemini Robotics ER 2 for improved video understanding and multi-robot task orchestration in physical automation projects.
• •MCP spec support in mcp-handler Vercel updated mcp-handler to version 2.0 with the latest Model Context Protocol spec so developers can run stateless MCP servers on Next.js and similar frameworks.
• •Gemini Robotics ER 2 details DeepMind published further benchmarks showing Gemini Robotics ER 2 gains in tool use and robot collaboration for real-world tasks.

Replicate new models

Replicate added five new models today. Two focus on video generation and text-to-image while the rest cover voice and LoRA tasks. All expose HTTP APIs for immediate calls from no-code stacks.

• •ltxvideo-2.3-lora Replicate released ltxvideo-2.3-lora for video generation with community LoRAs for camera control and ingredient prompts at low per-run cost.
• •p-image-ideogram Pruna AI released p-image-ideogram, a text-to-image model in collaboration with Ideogram that offers four quality tiers starting at $0.003 per generation.
• •voice-over-app-cloud Irulkenzei released voice-over-app-cloud for text-to-speech conversion with adjustable speed and language options via simple API calls.
• •p-image-ideogram: Pruna AI listed p-image-ideogram on Replicate with confirmed pricing tiers so Vibe Builders can test image output without local setup.
• •ltxvideo-2.3-lora: Replicate listed ltxvideo-2.3-lora highlighting LoRA support for quick video edits inside existing agent pipelines.

Product Hunt picks

Three AI-adjacent tools appeared on Product Hunt today. They target meeting notes, motion graphics, and keyboard-driven Mac control. Each offers direct paths for non-coders to test without heavy setup.

• •Sorinai Sorinai launched an interactive AI notepad that records and summarizes meetings for users who need quick transcripts without manual editing.
• •Premation Premation released an open-source AI tool positioned as an After Effects alternative for motion graphics work on a budget.

Other

AWS and Databricks published four posts on monitoring, migration, and agentic workflows. The pieces focus on production monitoring and data foundations rather than new model releases.

• •SageMaker meta-monitoring AWS added inference meta-monitoring for SageMaker endpoints through Amazon QuickSight dashboards for teams tracking model health at scale.
• •AI-forward healthcare foundations Databricks outlined steps for healthcare organizations to build reliable data layers before rolling out clinical AI tools.
• •Agentic media buying on Databricks Databricks described coordination layers needed for buyers and sellers to run agentic media campaigns without manual handoffs.

Industry news

Seven stories covered platform changes and market pressure. LinkedIn and Google addressed AI quality and bug fixes while OpenAI cut prices and researchers flagged data costs. Regulatory and wearable updates also appeared.

• •LinkedIn AI slop reporting LinkedIn added a report button for AI-generated low-quality posts and replaced its AI writing tool with a lighter proofreading option.
• •Friend wearable update Friend relaunched its AI wearable with voice output at a higher price point for users seeking always-on personal agents.
• •Google Chrome bug fixes Google credited AI tools for fixing more Chrome bugs in June than in the prior two years combined.
• •OpenAI GPT-5.6 price cuts OpenAI cut GPT-5.6 Luna prices 80 percent and Terra 20 percent citing efficiency gains and competition from lower-cost providers.
• •Training data investment forecast Former OpenAI researcher Andrew Ho predicts labs will spend over $100 billion on specialized training data as scaling alone plateaus.

What this means for you

For Vibe Builders: You can now test image-text and voice models like Inkling-Small and Audio8-TTS directly on Hugging Face or Replicate without writing code. Vercel Sandbox and AI Gateway updates let you run multiple agents side-by-side and pay less for GPT-5.6 calls. Product Hunt tools such as Sorinai give quick meeting-note agents you can try today.

For Non-techies: OpenAI price cuts and new Replicate models mean cheaper image and voice features for your daily tools. LinkedIn now lets you flag low-quality AI posts while Google improved Chrome stability with AI bug fixes. Databricks stories show how larger businesses are adding monitoring before scaling AI inside operations.

For Developers: Vercel added multi-user Sandbox isolation and MCP 2.0 support so you can run production agent fleets with private directories. Hugging Face trending models and Replicate LoRAs give immediate inference endpoints to benchmark against your current stack. OpenAI 80 percent Luna cut and Google bug-fix numbers signal you should retest cost and reliability assumptions this week.

What to watch next

Track Vercel Sandbox adoption metrics and any follow-up on the 80 percent GPT-5.6 price cut. Watch Replicate run counts for ltxvideo-2.3-lora and p-image-ideogram as signals of no-code uptake. Monitor LinkedIn policy changes for shifts in acceptable AI content volume.

Sources

Hugging Face trending

• •Inkling-Small by thinkingmachines trends on HuggingFace
• •Qwen3.6-35B-A3B-Escha-W2 by EschaLabs trends on HuggingFace
• •Audio8-TTS-Preview-0.6b by Audio8 trends on HuggingFace
• •Instella-MoE-16B-A3B-Think by amd trends on HuggingFace

Vendor launches

• •Gemini Spark now integrates with Chrome
• •Shopify and Vercel are rebuilding Hydrogen for faster storefronts
• •Inkling Small from Thinking Machines is now available on AI Gateway
• •Run multiple isolated agents in a single Sandbox
• •Deployments are now up to 7 seconds faster
• •AI Gateway: GPT-5.6 pricing and speed updates
• •Introducing Gemini Robotics ER 2
• •Latest MCP spec now supported in mcp-handler
• •Gemini Robotics ER 2: powering robotics with video understanding, task orchestration, and multi-robot collaboration
• •Best in Class: Stream PC Games and Study on the Same Laptop With GeForce NOW

Replicate new models

• •ltxvideo-2.3-lora by replicate launches on Replicate
• •p-image-ideogram by prunaai launches on Replicate
• •voice-over-app-cloud by irulkenzei launches on Replicate

Product Hunt picks

• •Sorinai
• •Premation
• •Cursor Crane

Other

• •Inference meta-monitoring for Amazon SageMaker AI endpoints with Amazon Quick
• •NBCUniversal’s Seamless Migration: Unlocking Scalable Analytics with Databricks
• •Foundations for an AI-forward healthcare organization
• •Agentic media buying cannot scale without the right foundation. See how buyers and sellers get there on Databricks.

Industry news

• •LinkedIn adds a button to report AI-generated ‘slop’
• •Friend, the lonely AI wearable, returns with a new voice and a much bigger price tag
• •Google says it fixed more Chrome bugs in June than over the past two years, thanks to AI
• •OpenAI goes full China pricing mode with an 80 percent cut to its most affordable GPT-5.6 model
• •Ex-OpenAI researcher bets $100 billion will flow into training data because scaling alone won't cut it
• •Judge says Trump admin still lacks evidence for Anthropic ‘supply-chain risk’ label
• •Everyone Is Freaking Out About OpenAI and Anthropic’s Race for Dominance