Welcome back! We are watching the biggest companies rewrite the baseline for what AI should cost and where it should live. Anthropic just gave its free users an upgrade that was previously reserved for expensive enterprise tiers. Perplexity is walking away from the advertising model to sell subscriptions directly to executives, citing major issues with user trust. Meanwhile, Apple is actively trying to move Siri out of your phone and into your clothing.
In today’s Generative AI Newsletter:
Anthropic launches Sonnet 4.6 with 1M token
Norton Neo: The safe AI browser that works for you
Nimbalyst: Built for faster AI agent iteration
Apple develops three new wearables for Siri
Perplexity drops its ad strategy to focus on subscription
MiniMax offers a fast, OpenAI-compatible API
Latest Developments
Claude Sonnet 4.6 Can Make Free Chat Handle Massive Text
Anthropic released Claude Sonnet 4.6 and set it as the default model for Free and Pro on claude, pitching it as a full upgrade without raising prices. The most concrete claim is that it can handle 1M tokens in beta of text in one go, reporting that people picked it over Sonnet 4.5, 70% of the time in early testing. If you use AI everyday, the tool now encourages you to paste more of your job into a single chat. You save time by keeping everything in one place, but one wrong reading can spread across the output.
Here’s what the product reveals:
Context: '1M tokens in beta' means you can include long reports or big chunks of code.
Pricing: Anthropic kept API pricing at $3/$15 per million tokens, so developers pay the same rates for the latest model.
Features: Context compaction shrinks long material so the model can keep working without re-uploading.
Safety: Anthropic claims 4.6 resists prompt injection better than the 4.5 model.
This is what the AI market looks like in 2026, where the best upgrades arrive disguised as defaults that change what people expect from a free tool. Sonnet 4.6 will help a lot of users move faster, especially on coding and document-heavy work. However, it can also amplify errors because a model that acts more can still choose the wrong action. The next phase will reward people who build habits around verification, especially when the model sounds certain.
World’s First Safe AI-Native Browser
AI should work for you, not the other way around. Norton Neo is the world's first safe AI-native browser with context-aware AI, built-in privacy, and configurable memory. Zero-prompt productivity that actually works.
Special highlight from our network
Managing agents requires a purpose-built tool to increase iteration speed and bandwidth.
Why Nimbalyst works better:
Interactive visual editing:
Work visually with Claude Code across editable markdown, mockups, diagrams, CSV files, Excalidraw, data models, MCP, and code. See every AI change. Approve what moves forward.
Parallel session management:
Clear, readable outputs. Organize, search, and resume sessions anytime. Link sessions to modified files with full context.
Developer mode:
Built-in Git management, commits, worktrees, and an embedded terminal. Everything in one place.
Apple Siri Could Learn What You See Using These New AI Devices
Apple is building Siri for the world outside your phone. Bloomberg reports that Apple has restarted work on three AI wearables: smart glasses, a camera pendant, and AirPods with cameras built to turn everyday scenes into prompts for Siri to understand what you are looking at. This lets Apple enable you to ask for help while you walk, shop, travel, or work, without pulling out your phone, but it also raises a social cost because people tend to distrust gadgets that might be watching them.
Here are the details:
Glasses: Apple targets 2027 and wants glasses with no display, leaning on cameras plus Siri to interpret the world.
AirPods: They come with small cameras that gather visual hints for Siri, and they may arrive as early as this year.
Pendant: Apple may be testing a pin or pendant that clips to clothing and uses a camera and microphones while the iPhone deals with the heavy lifting.
Privacy: These wearables could create friction at schools, offices, concerts, airports, and sensitive workplaces.
Apple is not selling a smarter chat box as much as a helper that notices and learns your day in real time. That future can feel genuinely useful and effortless for translation, navigation, daily tasks and reminders. It can also replay the AI Pin story, where the idea sounds useful with nicer hardware but the same social discomfort. Apple will need to prove that 'helpful' devices do not turn creepy while invading privacy.
Perplexity Turns ‘Ad-Free’ to Stop Them From Shaping Answers
Perplexity is changing how it plans to make money after reworking its business model to stop testing ads. As per recent reports, the company is stepping back from advertising and leaning into subscriptions and enterprise sales. It keeps a free version so anyone can try it, then limits heavy users and companies till they pay. Perplexity’s leaders describe ads as something that can hurt trust when they sit near the guidance.
Here are the specifics:
Retention: The internal scoreboard leans toward people coming back regularly, not just one-time spikes of curiosity traffic.
Money: It claims revenue grew 4.7x last year and reached about $200M in yearly recurring revenue by Oct 2025.
Enterprise: Its enterprise sales team is about five people now, with plans to hire more to sell to doctors and executives.
Strategy: The CEO said, “We underestimated people’s willingness to pay,” which matches the shift toward subscriptions.
Perplexity’s move also sounds like a warning to the rest of AI. Ads solved that problem for search and social media, but AI answers feel different because users treat them like advice. Perplexity is indicating that people will pay to avoid doubt at the exact moment they need certainty, which might turn these tools into premium services that leave casual users behind. If AI companies place ads near answers, they may win short-term revenue and lose the long-term habit.
MiniMax API Platform: OpenAI-Compatible Models for Text, Audio, and Video
MiniMax is an AI platform you can use through an API to do practical work: refactor code, extract structured data from long docs, generate voiceovers, and prototype short video clips. If you already use an OpenAI-style API, you can often plug MiniMax in quickly and start testing outputs without rebuilding your integration.
Core functions (and how to use them):
OpenAI-compatible requests: Point your existing OpenAI SDK/client to MiniMax, keep the same request shape, and rerun a few real prompts to compare output quality and latency.
Long-document extraction: Paste a PRD, policy doc, or transcript and ask for strict JSON (requirements, risks, decisions, action items). Use that JSON to populate Notion/Jira without manual cleanup.
Code refactor + tests: Give it a function and ask for a patch-style diff plus pytest tests. Apply the diff, run tests, then ask for a second pass only on what fails.
Text-to-speech voiceovers: Convert a script into audio for demos or training. Provide pronunciations for brand terms and request two takes (neutral + upbeat) so you can choose fast.
Video + narration drafts: Generate a short video concept from a brief, then generate matching narration audio from the same brief so your draft stays consistent.
Try this yourself:
Take a document you already have (a PRD, a meeting transcript, or a long Slack thread) and turn it into something you can paste directly into your workflow. Run the same prompt on two different docs and compare how consistently it keeps the structure once it’s reliable, you’ve got a reusable “converter” prompt you can use every week.
