What Happened This Week
1. Claude's New Model: Better Coding, Cheaper Price
Anthropic released its latest model: Claude Opus 4.5.
November has been loaded with new model releases.
This was this week's big drop.
It's Anthropic's new top-tier model for reasoning, coding, and agents.
For coding benchmarks, it surpassed GPT-5.1-Codex-Max from OpenAI (which only came out a week ago).
This release is an important move.
Anthropic has had the go-to coding models for most of the past year.
OpenAI has been chipping away at Anthropic's main differentiator.
That's gotta be concerning.
This puts them back in the top spot for coding capabilities for now.
The other major factor is price.
OpenAI's GPT-5.1-Codex-Max model is cheap - only $1.25 per 1M input tokens.
Comparatively, Opus 4.5 is $5 per 1M input tokens.
Still costlier than Codex-Max, but significantly cheaper than Claude's Opus 4.1 model which cost $15 per 1M input tokens.
Here's what people are saying Opus 4.5 is good and not so good at:
| What Opus 4.5 Is Good At | What Opus 4.5 Struggles With |
|---|---|
| Handling messy or complex work where you need to think through tradeoffs and options. | Being a “tiny tasks” bot for simple things like quick rewrites and general support where a lighter model is enough. |
| Staying on top of big, long projects (plans, research, multi-step tasks) without forgetting what you said earlier. | Always giving short, punchy answers; it can drift into longer explanations than you really need. |
| Turning ideas into actionable work: outlines, project plans, meeting agendas, checklists, briefings. | Writing with a very distinct brand voice or heavy emotional tone compared with models tuned just for creative copy. |
| Helping you decide what to do: comparing options, spotting risks, and suggesting better ways to run a process. | Following rigid templates perfectly every time; it still needs a quick skim before sending to stakeholders. |
| Acting as an operations assistant that can remember context, update plans, and help manage ongoing work. | Cost and power can be overkill for simple work; best to save it for your harder, more important tasks. |
The signal?
Your engineering teams keep getting better, specialized models that cost less.
This means capabilities keep improving and development cycles speed up.
Majority of non-technical workers won't use this model.
If you're interested in trying it, test it for a complex task that needs reasoning.
Either way, you'll feel the downstream effects of continued coding improvements.
2. ChatGPT Launches Shopping Research Tool
OpenAI launched shopping research in ChatGPT.
It'll help you shop by providing product guides, comparisons, and recos.
It's rolling out now across plans with nearly unlimited usage through the holidays.
To use it, select "Shopping research" from the (+) menu in ChatGPT.
Or it'll suggest it automatically when you ask shopping questions.
Describe what you need (ex: "help me find a gift for my office holiday exchange.")
After it'll ask you questions about budget and preferences.
Then it gives initial recos that you can mark as "not interested" or "more like this" to steer additional recommendations.
In a few minutes, you get a personalized guide with products, differences, and tradeoffs.
The signal?
Last week, we talked about Google and Target's AI shopping moves.
This week, more commerce is coming to your AI experiences.
Once again, for brands it's crucial to have high-quality product data.
Clean specs, clear descriptions, accurate pricing, and robust reviews.
Messy data will keep you invisible and you'll miss out on potential customers.
This is also a signal of where AI monetization is going.
AI orgs could get revenue through promoted products or through affiliate models that give them a cut of purchases they drive.
Worth trying this and Google's upgrades if you're shopping this season.
3. Perplexity Launches Shopping with PayPal Integration
Perplexity launched AI-powered shopping this week with PayPal.
It lets you research products and buy them inside Perplexity using PayPal "Instant Buy" button.
Perplexity is the third major AI platform to roll out shopping features this month.
Here's how Google, OpenAI, and Perplexity features compare:
| Feature | Google AI Shopping | ChatGPT Shopping Research | Perplexity + PayPal |
|---|---|---|---|
| What it does | AI shopping in Search and Gemini with product comparisons and recommendations | Deep buyer's guides with research, tradeoffs, and personalized recommendations | Conversational search that surfaces products you can buy directly in-app |
| How to purchase | Sends you to retailers (Google Pay checkout available for select merchants) | Sends you to retailers (Instant Checkout available for select merchants) | Buy directly in Perplexity via PayPal Instant Buy, no leaving the app |
| Best for | Quick product discovery and price comparisons while searching | Deep research when comparing multiple options with specific constraints | Fast search-to-purchase flow, especially if you already use PayPal |
The signal?
Shopping features are becoming standard across AI platforms.
Just like deep research tools or image gen, shopping will be table stakes too.
The timing for these releases is important too.
All three rolled out shopping right before the holidays for peak shopping season.
4. ChatGPT Group Chats Roll Out Globally
After testing them in a handful of markets, OpenAI officially rolled the feature out worldwide.
Now up to 20 people can collaborate in a single conversation with AI.
Here's how it works:
Click the people icon in ChatGPT and invite others with a link.
Everyone can type in the conversation and interact with ChatGPT together.
Use cases: team brainstorming, project kickoffs, collaborative planning, decision-making sessions.
Privacy controls keep group chats separate from your private threads.
It's rolling out to Free, Go, Plus, and Pro users globally.
Enterprise and business plans coming soon.
The signal?
This is where AI collaboration is headed.
It'll have to shift from solo assistant to a collaborative team member to truly integrate into the workforce.
But the execution has practical challenges.
Today's AI outputs are long.
Group texts are already hard to follow when multiple people are talking.
Add essay-length AI responses and it gets unbearable fast.
For this to work well, AI needs to be a lot briefer and talk more like a person in group settings.
For now, this feature is worth trying, since that is the direction things are moving.
That said, expect some rough edges in these early days.
5. Numbers Worth Knowing This Week
Two stats that tell you where things are headed:
- Perplexity's downloads dropped 80% in six weeks.
- Part of this is a promo surge cooling off, about 50% of downloads since July came from a partnership offering free Pro subscriptions in India.
- A bigger factor is competition. ChatGPT and Google keep getting better at what Perplexity stood out for: fetching and analyzing live web results.
- The signal? Differentiators erode fast these days. When competitors can replicate a core feature, the advantage disappears quickly.
- Anthropic analyzed 100,000 Claude conversations and found AI cuts task completion time by roughly 80%.
- Widespread adoption could boost annual U.S. labor productivity growth by 1.8% (doubling the current rate).
- Top roles seeing gains: software developers (19%), ops managers, marketing specialists.
- Biggest time savings: curriculum development (96%), research assistance (91%), executive admin (87%).
- The signal? The productivity gains are measurable, but impact varies widely. Some tasks see massive speedups, in others AI is a bottleneck.
