ContentBuffer
New research suggests open source AI isn't cannibalizing the market for expensive state-of-the-art models. Despite the rise of lighter alternatives, frontier labs like Anthropic maintain their position by dominating early-stage use cases. Meanwhile, DeepSeek has surged into first place in token volumes on Vercel's platform, with OpenRouter processing 5.3 trillion tokens weekly using DeepSeek V4 Flash at a fraction of the cost compared to Opus 4.8 ($1.37 per million tokens vs 6 cents).
Microsoft has started deploying its own MAI models to cut costs within Office 365, reducing dependence on external providers like OpenAI and Anthropic. Developers should care because this trend signals a broader industry move towards cost optimization in AI services. Microsoft's new approach affects the economics of integrating third-party AI into enterprise software, potentially shifting market dynamics. The company launched seven MAI models at Build 2023, including an agentic coder and text-to-image generator. Let's dive deeper…
In today's ContentBuffer update:
Discord's AI moderation bug mistakenly bans over 8,000 accounts
Google to Unveil Pixel 11 with Design Upgrades on Aug 12
Figma Expands Into AI with Bud Buyout
Netflix Launches Live, Game, and Podcast Content on Platform
5 new AI tools & 5 new AI jobs
More tech news
Latest Development
Image source: techcrunch.com
Summary: Frontier models like Anthropic are holding steady despite open source AI's growth, thanks to their dominance in early-stage use cases.
Key Points:
DeepSeek leads Vercel's AI gateway dashboard with the highest token volumes at 5.3 trillion tokens weekly.
Z.ai jumps into fourth place on the platform, showing strong adoption of its services.
Anthropic accounts for over half of all AI spend on Vercel’s platform, maintaining a dominant position.
OpenRouter captures a larger market segment with DeepSeek V4 Flash processing 5.3 trillion tokens weekly at a lower cost.
Opus 4.8 handles just over 2 trillion tokens weekly but costs roughly 23x more than V4 Flash.
Why it matters: If you're working on early-stage AI projects, frontier models like Anthropic's still offer unmatched capabilities despite higher costs. As these use cases mature, cheaper open-source alternatives can take over, maintaining a two-tiered model economy.
Image source: techcrunch.com
Summary: Microsoft is cutting costs by using its own MAI models in Office 365 instead of relying on OpenAI and Anthropic — a move that signals broader industry cost-cutting trends.
Key Points:
Microsoft launched seven new MAI models at Build 2023, including an agentic coder and text-to-image generator.
The company is using its in-house AI to respond to user prompts in Excel and Word, reducing reliance on external software providers.
Previously, large parts of Office 365 were powered by models from OpenAI and Anthropic; now MAI handles a significant portion.
Microsoft's cost-cutting efforts are part of a broader industry trend as companies like Amazon, Uber, Meta, and Accenture curb spending.
Some firms look to Chinese AI models for cheaper agentic solutions despite security concerns.
Why it matters: If you're building enterprise software with integrated AI, Microsoft's shift to in-house MAI models impacts the economics of third-party reliance. For instance, a company using OpenAI or Anthropic might see competitive pressure from those offering similar services at lower costs due to internal development.
Image source: techcrunch.com
Summary: Discord's AI moderation glitch mistakenly banned over 8K users since May, bypassing human review. All accounts are being restored as the company works on better safeguards.
Key Points:
A bug in Discord’s AI moderation system caused wrongful bans of over 8,000 user accounts since May.
An additional 200 users were banned over the weekend before the issue was fixed.
Discord's automated safety system matches uploaded content against databases of known harmful material.
Human moderators usually review flagged content before any action is taken; this bug skipped that step.
The incident has sparked calls for increased transparency in AI-assisted moderation practices.
Why it matters: If you're using Discord or similar platforms with automated safety systems, the false positives from AI can lead to wrongful bans. This affects user trust and highlights the need for better safeguards against such issues.
Image source: techcrunch.com
Summary: Google to unveil Pixel 11 lineup at August 12 NYC event — design upgrades, new gold color variant, and higher storage minimums.
Key Points:
Pixel 11 launch scheduled for August 12 in New York City
Base model may feature slimmer bezels and a sleek black camera bar
Rumored new gold color variant for Pixel 11 lineup
Pixel 11 Pro expected to be slightly thinner than its predecessor
Google may skip the 128GB option, starting with 256GB storage
Why it matters: If you're in the market for a new Android phone, the Pixel 11's design upgrades and higher minimum storage could impact your budget. The gold color variant might attract fashion-conscious buyers, but the price hike due to skipped lower-tier storage options may deter some users.
Image source: techcrunch.com
Summary: Figma acquires Bud's team, expanding its AI toolkit for app development. Developers should pay attention as this could change how they build and prototype apps.
Key Points:
Bud allows users to create apps for mobile, web, Slack, and browser platforms.
The acquisition aims to give Figma teams more tools for building and prototyping apps with AI.
Figma previously released 'Make' last year for creating web applications.
Users of Bud need to migrate their projects by July 18th as the platform shuts down.
Bud's AI agent can access various services, browse the web, and write code.
Why it matters: If you're using Figma to build or prototype apps, this acquisition means more tools for automation and cross-platform development. The shutdown of Bud by July 18th forces users to migrate their projects, impacting current workflows. This move positions Figma as a leader in integrating AI into the full-stack development process.
Image source: techcrunch.com
Summary: Netflix is testing live content, video games, and podcasts from publishers like BuzzFeed Studios and Condé Nast. Available August 3 in select markets.
Key Points:
New formats include live broadcasts, video games, and video podcasts from multiple publishers
Available starting August 3 in the US, Canada, UK, Ireland, Australia, and New Zealand
Content ranges from two to over 20 minutes long, catering to short-form trends
Netflix aims to test audience appetite for these formats before scaling production
Partnerships include BuzzFeed Studios, Condé Nast, Hearst Magazines, People Inc., Tastemade
Why it matters: If you're a Netflix subscriber in select markets, prepare for new content types. The move to short-form and interactive media signals a shift towards competing with platforms like YouTube and TikTok.
New Tools & Job
Sakana Marlin - Sakana Marlin is Sakana AI's first commercial product: an autonomous 'Ultra Deep Research' agent positioned as a Virtual Chief Strategy Officer. Given a research topic, Marlin works autonomously for up to roughly eight hours, forming hypotheses, gathering and reconciling information from online sources, and synthesizing a detailed strategy report up to ~100 pages plus executive-summary slides. It builds on Sakana AI research including AB-MCTS (NeurIPS 2025 Spotlight) and The AI Scientist (published in Nature). A closed beta ran from April 2026 with ~300 professionals across finance and consulting.
Upsolve AI - Upsolve AI is an Agent Studio for data teams to build, deploy, and evaluate grounded, governed, and trustworthy analytics agents. It layers institutional context (warehouse tables, validated SQL patterns, semantic models, KPI definitions, and business rules from sources like Notion, Slack, and email) on top of 30+ SQL database connectors so end users can ask questions in natural language and get accurate, traceable answers. Every conversation is fully traced end-to-end (tool calls, SQL, agent output), a built-in evaluation agent grades performance, and context monitoring surfaces gaps so accuracy improves with use. Agents deploy to Slack, Teams, Claude, ChatGPT, Cursor, or embed in your own product via MCP/SDK.
Buddy - Buddy is an AI design agent that lives in the Figma canvas. Chat to generate screens, flows, components, and variants using your existing design system; paste a public URL or drop an HTML file to recreate live websites as editable Figma layers; and run batch edits and layer organization across files. Supports multiple LLMs (Opus, Sonnet, GPT, Gemini, Codex). Works on any Figma plan including Free.
Tyto - Tyto by ai-coustics is a monitoring and diagnostics layer for Voice AI. It produces a single real-time risk score that predicts speech-to-text, voice-activity-detection, and turn-taking failures before they occur, scoring six qualities of incoming audio so teams can catch audio issues before they impact voice agents in production.
Adapt - Adapt is one shared context layer for every employee and every agent. It connects to the tools a company runs on, holds live context, answers questions, and takes action that does the work and cites its sources, in Slack and on the web. Data stays private and is never used to train models.
Machine Learning Solutions Engineer – Robotics - The next frontier for AI is the physical world. At Scale, we're pioneering this shift, moving artificial intelligence from digital spaces into robotics. Our Robotics team builds the critical infrastructure that empowers the most sophisticat…
Solutions Architect, AI/Cloudflare Developer Platform - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Interne…
Senior Manager, Solutions Architecture, AI & Developer Platform - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Interne…
People Team Intern - HR Operations & AI Innovation (Fall 2026) - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Interne…
Head of GTM, AI Inference - About Us At Cloudflare, we are on a mission to help build a better Internet. Today the company runs one of the world’s largest networks that powers millions of websites and other Interne…
QUICK HITS
Cyber Attacks Surge: Grids, Water Plants Targeted - Nation-state hackers are ramping up destructive cyberattacks on civilian infrastructure, putting real-world safety at risk. This is a wake-up call for better security practices.
Claude Cowork Expands to Web and Mobile for General Work - Anthropic launched Claude Cowork on web and mobile, aiming to make AI a constant presence in daily workflows beyond coding tasks. The move is significant for anyone looking to integrate AI into their day-to-day work.
X Adds Video Editing Tools for Original Content - X introduces video editing tools in its iOS app to boost original content creation. Why you should care: It’s a step towards curbing recycled material on the platform.
Chemistry Ventures Raises $500M for Oversubscribed Second Fund - Chemistry Ventures is raising a whopping $500M for its oversubscribed second fund, targeting early-stage startups in fintech and developer tools. More money means more opportunities for founders to grow their businesses.
Norm Secures $120M Series C for AI Law Firm - Norm just raised $120M for their AI law firm, aiming to disrupt traditional legal billing with outcome-based pricing. Why you should care: If you're an enterprise client looking for innovative legal solutions, this could be a game changer.
Bidbus Raises $15M to Beat Carvana's Prices by $2K-$3K - Bidbus raised $15M to outbid Carvana by up to $3K per car. Dealers compete for your ride with higher offers on their platform.
THAT'S A WRAP
Our newsletter reaches a diverse audience of tech professionals, investors, engineers, managers, and business leaders worldwide. Get in touch today.
Be the first to comment
Get the daily edition delivered free every morning.
Microsoft is using in-house MAI models to reduce costs, cutting reliance on OpenAI and Anthropic. This shift impacts the economics of AI integration in enterprise software.
Apple introduces voice customization in Siri beta, allowing users to adjust pace and expressiveness. This mirrors features found in competitors like ChatGPT.
Starting July 10, Alibaba employees are banned from using Anthropic's Claude Code due to security concerns. The company views the software as high-risk and is promoting its own Qoder tool instead.