DeepSeek V4 Flash 90% off, NVIDIA Alpamayo 2 for AVs, and agent tools shipping today
TL;DR
Vendors pushed model discounts, autonomous vehicle models, faster deploys, and browser-equipped agents while open models and cloud deals advanced across the board.
What shipped
On 4 August 2026 vendors released discounted inference, new open models for vehicles, faster static regeneration, and sandboxed agent browsers. NVIDIA led with multiple infrastructure and safety updates while Vercel and others shipped developer-facing speed and tooling gains. Hugging Face and Replicate added trending models and video generation options.
Vendor launches
NVIDIA released the largest share of updates with an open model for autonomous vehicles, storage architecture guidance, education hub participation, and cybersecurity guidelines. Vercel followed with a DeepSeek discount, faster ISR deploys, and an agent browser extension. Convex announced funding while Anthropic and Google added leadership and monthly recap notes.
- •DeepSeek V4 Flash discount Vercel made DeepSeek V4 Flash available at 90 percent off through Novita on AI Gateway for Pro customers until August 11, with fallback to standard rates afterward.
- •NVIDIA storage guidance NVIDIA explained how AI context windows now exceed system memory and require purpose-built storage architectures that deliver grounded insights from large datasets.
- •NVIDIA Alpamayo 2 model NVIDIA released Alpamayo 2 Super, an open model for robotaxis and autonomous vehicles that reasons about rare long-tail events beyond basic detection.
- •Vercel ISR deploy speed Vercel cut deploy times up to 33 percent for apps with many ISR pages by bundling routing metadata instead of splitting uploads.
- •SAFE cybersecurity guidelines The Open Secure AI Alliance released a request for comments on Shared AI Findings Exchange to improve transparency for agentic AI systems.
- •Google July recap Google published a summary of its July 2026 AI releases covering model and product updates from the prior month.
- •Convex Series B funding Convex raised $57 million to build backends that support developers and agents writing software together.
- •Eve agent browser extension Vercel released an agent-browser extension that lets eve agents navigate pages, fill forms, take screenshots, and inspect network activity inside a sandbox.
Hugging Face trending
Four models trended on Hugging Face Hub. LiquidAI and LGAI-EXAONE each placed text-generation models, while realrebelai and ethanfel contributed a text-to-video model and an image-text-to-text pipeline respectively.
- •LFM2.5-2.6B model LiquidAI placed LFM2.5-2.6B, a 2.6 billion parameter text-generation model, on the Hub for download and fine-tuning.
- •MiniMax-H3_GGUFs model realrebelai uploaded MiniMax-H3_GGUFs, a text-to-video model available for inference and fine-tuning on the Hub.
- •K-EXAONE-2.0-750B model LGAI-EXAONE released K-EXAONE-2.0-750B-A37B, a 750 billion parameter text-generation model built with transformers.
- •Qwen3-VL pipeline ethanfel shared Qwen3-VL-32B-Ultra-Heretic-MiniMax-H3-ComfyUI-INT8-ConvRot, an image-text-to-text model using ComfyUI.
Replicate new models
Black Forest Labs added one new model to Replicate. The early-access FLUX 3 preview supports video generation with synchronized audio from text, images, or video inputs.
- •FLUX 3 on Replicate Black Forest Labs launched FLUX 3, a multimodal model that generates video with synchronized audio from text, images, or video.
- •FLUX 3 early access The same FLUX 3 model became available for direct HTTP API calls or playground testing on Replicate.
Product Hunt picks
Five AI tools appeared on Product Hunt. They target outbound sales, note search inside chat interfaces, hardware for coding, Claude Code terminals, and governed workflow languages.
- •Stynar SDR tool Stynar launched an AI sales development representative that runs outbound campaigns without manual setup.
- •Glasp MCP Connector Glasp released a connector that lets users search their highlights and notes inside Claude and ChatGPT.
- •Vibe Buddy hardware Vibe Buddy introduced hardware designed to assist AI coding workflows.
- •MOTHER terminal MOTHER shipped a terminal for Claude Code that supports one-click session resume.
- •Finyuus language Finyuus introduced a code-first language for building durable and governed AI workflows.
Other
AWS, Black Forest Labs, GitHub, and Databricks each released one update. The items cover web insight agents, FLUX 3 video research, stacked pull requests for AI code, web search grounding, and an AI gateway.
- •Bedrock AgentCore extraction AWS released automated web insight extraction using Amazon Bedrock AgentCore for structured data collection.
- •FLUX 3 video research Black Forest Labs published the first part of its FLUX 3 video generation research covering core generation techniques.
- •GitHub stacked PRs GitHub showed how to break one large AI-generated pull request into an ordered, reviewable stack of smaller PRs.
- •Bedrock web search AWS added web search to Amazon Bedrock to ground foundation model responses with current external data.
- •Unity AI Gateway GA Databricks made Unity AI Gateway generally available to manage AI model access and governance.
Industry news
Five stories covered AI adoption and business moves. Pulitzer winners disclosed record AI use, Google structured Anthropic chip financing, open-weight models advanced, Anthropic signed a large cloud deal, and the Open Secure AI Alliance released early proposals.
- •Pulitzer AI disclosures Eight Pulitzer Prize entries, including five winners, disclosed AI use for document search while keeping writing and editing human-only.
- •Google Anthropic financing Google arranged multibillion-dollar financing with Broadcom and others to supply Anthropic chips and data centers off its balance sheet.
- •Open-weight model safety A SaferAI report found Z.ai GLM-5.2 near frontier capability yet missing key safety mitigations compared with closed models.
- •Anthropic Volta deal Anthropic signed a reported $10 billion cloud agreement with startup Volta for additional AI infrastructure.
- •Open Secure AI Alliance progress The week-old alliance led by NVIDIA and now over 120 members released initial proposals for defending against agentic AI threats.
What this means for you
For Vibe Builders: You can now route DeepSeek V4 Flash at heavy discount through Vercel AI Gateway and give eve agents a sandboxed browser for web tasks. FLUX 3 on Replicate and Product Hunt tools like Stynar and Vibe Buddy let you test video generation and outbound automation without writing infrastructure code. Watch Convex backend updates if your agents need durable state.
For Non-techies: AI tools moved from chat to action with browser-equipped agents and sales SDRs like Stynar that run campaigns for you. Note search connectors and one-click terminals reduce setup time for daily workflows. Record AI use in newsrooms shows the technology is already embedded in professional output.
For Developers: Vercel cut ISR deploy times 33 percent and added agent browser tools while Convex raised funding for agent-friendly backends. GitHub guidance on stacked PRs and AWS web search grounding give concrete ways to productionize agent output. Track the Open Secure AI Alliance proposals for upcoming agent security requirements.
What to watch next
Watch Anthropic Volta deployment details and any early SAFE guideline comments. Track whether FLUX 3 video quality holds in production runs and whether open-weight models close the safety gap in the next SaferAI report.
Harsh’s take
The day showed clear separation between infrastructure vendors shipping measurable speed and cost wins and frontier labs focused on financing and policy positioning. NVIDIA dominated volume while smaller teams delivered the most immediately usable agent extensions. The practical gap remains between model releases and reliable, governed deployment. Builders should pick one new tool from the agent-browser or stacked-PR guidance and integrate it into an existing workflow this week to test real throughput gains.
by Harsh Desai
Sources
Vendor launches
- •DeepSeek V4 Flash is 90% off through Novita on AI Gateway
- •As AI Increases Demands on Memory, Storage Steps Up
- •NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available for Commercial Use
- •NVIDIA Joins NSF State and Regional AI Hubs Program to Expand AI Research and Education Across the US
- •Deploys are now up to 33% faster for apps with many ISR pages
- •AI Leaders Propose SAFE Guidelines for Cybersecurity Transparency
- •The latest AI news we announced in July 2026
- •Convex raises a $57M Series B
- •Give your eve agent a browser
- •Mariano-Florentino (Tino) Cuéllar to join Anthropic as Chief Global Affairs Officer
Hugging Face trending
- •LFM2.5-2.6B by LiquidAI trends on HuggingFace
- •MiniMax-H3_GGUFs by realrebelai trends on HuggingFace
- •K-EXAONE-2.0-750B-A37B by LGAI-EXAONE trends on HuggingFace
- •Qwen3-VL-32B-Ultra-Heretic-MiniMax-H3-ComfyUI-INT8-ConvRot by ethanfel trends on HuggingFace
Replicate new models
Product Hunt picks
Other
- •Automated web insight extraction with Amazon Bedrock AgentCore
- •ModelsResearchFLUX 3 Video, Part 1: Generation
- •Turn one giant AI-generated pull request to a reviewable stack
- •Introducing Web Search on Amazon Bedrock for foundation model grounding
- •Unity AI Gateway is Generally Available
Industry news
- •This year's Pulitzer Prizes saw a record number of winners disclose AI use
- •Google moves billions in Anthropic chip risk off its balance sheet
- •Open-weight AI models are catching up to the frontier. The safety gap remains.
- •Anthropic signs $10 billion deal with AI cloud startup Volta
- •Nvidia doesn’t mess around: A week after open AI industry group formed, it’s already showing progress
More AI news
- Daily RoundupQwen 3.8 debut, lipsync model on Replicate, and agent tools for daily builds
Vendors released new models and infrastructure updates while builders gained fresh agent and coding tools across Replicate, Hugging Face, and Product Hunt on 3 August.
- Weekly DigestHermes Agent v0.19.1 with GPT-5.6 and 80% latency cut, OpenClaw MCP apps (agent routing and recovery tools)
Hermes Agent rolled out v0.19.1 with new model support, durable ledgers, secret managers, and major speed gains while OpenClaw added MCP apps, structured questions, and local inference options across the same period.
- Daily RoundupCadence-fast on Replicate, Qwythos-27B on HF, and voice agents ready today
Fresh models appeared on Replicate and Hugging Face while Product Hunt surfaced local and voice agent tools, and Altman urged slower AI progress.