Skip to content
bge-m3 and websitescraper2 on Replicate, Qwopus3.6 trends on HF, Vercel agent tools (new models plus deployment updates) | Daily AI roundup cover

bge-m3 and websitescraper2 on Replicate, Qwopus3.6 trends on HF, Vercel agent tools (new models plus deployment updates)

By Harsh Desai
Share

TL;DR

On 1 July new embedding and scraping models landed on Replicate while five models trended on Hugging Face; Vercel shipped dry-run CLI checks, konsistent linting, Resend marketplace access and a security dashboard in beta.

What shipped

On 1 July model hosts and deployment platforms released fresh tools for inference and agent workflows. Replicate added two new models while Hugging Face showed five trending entries. Vercel and Google updates focused on preview checks, code consistency and marketplace integrations.

Replicate new models

Replicate added an embedding model and a web scraper that run via HTTP API. The releases give builders direct access to text processing and data extraction without managing servers. Both use Cog 0.21.0 and show usage counts in the hundreds or thousands.

  • bge-m3 by ifaas-uk ifaas-uk released bge-m3 on Replicate for text embedding work. The model accepts text or JSON lists and runs inference through the platform API. Vibe builders can test semantic search flows in the web playground without extra setup.
  • websitescraper2 by myaiteam2 myaiteam2 released websitescraper2 on Replicate to pull emails, phones, links and page content. The tool has logged over six thousand runs and supports direct API calls. SMB owners can extract contact data from sites for outreach lists.

Hugging Face trending

Five models gained traction on Hugging Face Hub, spanning code, text generation and panoramic image work. Most support transformers or Model Optimizer and list evaluation numbers in their cards. Builders can download or fine-tune them for local or cloud runs.

  • Qwopus3.6-35B-A3B-Coder-MTP-GGUF Jackrong released Qwopus3.6-35B-A3B-Coder-MTP-GGUF, a code-focused image-text-to-text model trending on the Hub. It ships with evaluation results for developers comparing it against other 35B variants.
  • Agents-A1 InternScience released Agents-A1, a text-generation model trending on the Hub. The entry includes usage examples that let vibe builders prototype simple agent loops without writing new training code.
  • Qwen3.6-27B-NVFP4 NVIDIA released Qwen3.6-27B-NVFP4, a text-generation model optimized with Model Optimizer and trending on the Hub. Engineers can benchmark it against earlier Qwen releases for throughput on production workloads.
  • Ornith-1.0-35B deepreinforce-ai released Ornith-1.0-35B, a text-generation model trending on the Hub. The card shows fine-tuning guidance that basic users can follow to adapt the model for domain-specific chat tasks.
  • SpheRoPE Researchers released SpheRoPE, a zero-shot framework that adds spherical priors to diffusion transformers for 360-degree images and videos. The paper reports no extra training or optimization steps required.

Vendor launches

Vercel and Google posted updates on CLI previews, code linting, marketplace additions and restored model access. The changes target faster agent testing and consistent code across human and machine edits. Several items include concrete commands or JSON output examples.

  • Google June 2026 AI updates Google published its June 2026 AI roundup covering model and platform changes. The post lists releases that developers can track for new API features.
  • B3 Android enterprise case B3 used Android and Samsung tools to scale mobile device management with AI features. The story shows how an SMB rolled out secure AI productivity tools to its workforce.
  • Vercel CLI dry-run Vercel added a dry-run flag to preview deployment files and framework detection before pushing code. Agents and builders can pipe the JSON manifest into automation scripts.
  • konsistent linter Vercel open-sourced konsistent, a CLI linter that enforces TypeScript folder and export patterns for agents and humans. Projects configure rules in a single JSON file.
  • Claude Fable 5 on AI Gateway Vercel restored Claude Fable 5 access after export controls lifted. The update includes stronger safety classifiers that still allow routine coding tasks.
  • Resend on Vercel Marketplace Resend joined the Vercel Marketplace so teams can send transactional emails from React components without managing SMTP. Delivery tracking works in real time.
  • Vercel Security Dashboard beta Vercel opened a private beta dashboard that flags missing 2FA, public previews and long-lived credentials across projects. Teams can review posture before agents spin up new environments.

Product Hunt picks

Six AI tools appeared on Product Hunt covering CAD, research, filmmaking, notebooks and video generation. Each targets a specific workflow from design to content creation. Builders can try the listed demos to test fit with existing stacks.

  • Adam CAD Copilot Adam CAD Copilot added AI assistance inside Onshape and Fusion for design tasks. Users can generate and edit CAD models without leaving the host application.
  • Claude Science Claude Science launched as a research partner that handles rigorous science queries and literature checks. The desktop version supports step-by-step reasoning traces.
  • Ciaro Pro Ciaro Pro released an AI filmmaking suite aimed at visual storytellers. It handles script-to-video pipelines with editing controls.
  • Clusy Clusy introduced an AI notebook platform built for modern data science workflows. Teams can run experiments and share reproducible notebooks in one place.
  • Claude Sonnet 5 Claude Sonnet 5 reached Product Hunt as an agent that plans and executes multi-step work. Early testers report it handles coding and research loops end to end.
  • Gemini Omni Flash Gemini Omni Flash added high-quality video generation with conversational editing controls. Creators can iterate on clips through chat prompts.

Other

Posts covered rapid product builds, protein design on SageMaker, open model hosting on Bedrock and automated verification steps. Several entries focus on moving AI from prototype to repeatable production flows. Builders can review the timelines and benchmarks shared.

  • Nursa 48-hour build Lovable shared how Nursa shipped a new product in 48 hours using AI tools and changed internal shipping habits. The case shows a full company process shift.
  • Camtasia alternatives review HeyGen published a comparison of 15 Camtasia alternatives tested for video editing speed and output quality. SMB owners can use the scores to pick a replacement.
  • BoltzGen on SageMaker AWS detailed how BoltzGen accelerates protein design workloads on SageMaker AI. Researchers can launch the model with managed GPU instances.
  • Nemotron and GPT OSS on Bedrock AWS added NVIDIA Nemotron and OpenAI GPT OSS models to Amazon Bedrock in GovCloud. Government teams can run the models under existing compliance controls.
  • Serverless A2A gateway AWS described a serverless gateway for agent discovery, routing and access control. The pattern lets builders register agents without standing up extra servers.
  • Augment pre-merge automation Augment Code automated pre-merge verification to catch issues before human review. The change reduced manual checks in their TypeScript codebase.
  • Qualcomm AI deployment view Qualcomm discussed moving AI from initial demos to sustained production use. The post highlights infrastructure and process gaps that stall rollouts.
  • Augment verification follow-up Augment Code followed up on verification automation and showed how the team handled the next bottleneck after initial fixes. The update includes metrics on pull request cycle time.

Industry news

Cloudflare set a September deadline for AI crawlers, SpaceX showed an AI device prototype and new funds formed around AI infrastructure. A reporting site for AI safety issues also launched. The items point to tighter data rules and hardware expansion.

  • Cloudflare crawler policy Cloudflare gave AI companies until 15 September to separate search crawlers from training and agent crawlers or face blocks on publisher sites. The change pressures firms to license content.
  • SpaceX AI device prototype SpaceX showed investors a handset-like AI device ahead of a potential public launch. The prototype signals expansion into wireless AI hardware.

What this means for you

For Vibe Builders: New Replicate models and trending Hugging Face entries give you ready-to-call endpoints for embeddings, scraping and code generation. Vercel dry-run CLI and konsistent linter let you test agent deployments and keep code consistent without writing extra scripts. Product Hunt tools such as Claude Sonnet 5 and Ciaro Pro offer quick trials for video and research workflows you can plug into existing projects this week.

For Non-techies: Replicate scrapers and email tools plus Resend marketplace access mean you can pull contact data and send messages from one dashboard. Google and Vercel updates on security checks and restored model access reduce the chance of sudden outages in daily AI tasks. Product Hunt picks like Claude Science and Gemini Omni Flash provide simple entry points for research notes or short video edits.

For Developers: Hugging Face trending models and AWS Bedrock additions supply concrete benchmarks for text and agent workloads you can compare against current stacks. Vercel CLI dry-run output and the new security dashboard surface misconfigurations before agents create new projects. Cloudflare policy and Augment verification posts flag upcoming crawler rules and pre-merge automation patterns worth testing in production pipelines.

What to watch next

Track whether Replicate usage for bge-m3 and websitescraper2 climbs past ten thousand runs. Watch for follow-up posts on konsistent rule examples and the Vercel Security Dashboard public beta date. Note any new model cards that cite the SpheRoPE paper for panoramic generation.

Harshs take

The day shows a split between quick-to-call models on Replicate and Hugging Face and deeper deployment hygiene work at Vercel. Most releases still require users to stitch together API keys, lint rules and safety classifiers rather than delivering one integrated flow. The practical move is to pick one new endpoint, such as bge-m3 or the dry-run CLI flag, and run it against an existing project before the next round of policy changes hits.

by Harsh Desai

Sources

Replicate new models

Hugging Face trending

Vendor launches

Product Hunt picks

Other

Industry news

More AI news

Everything AI. One email.
Every Monday.

New tools. Model launches. Plugins. Repos. Tactics. The moves the sharpest builders are making right now, before everyone else.

No spam. Unsubscribe anytime.