🎉 Perplexity Competes with Claude, New ChatGPT Model, Pew Study: Teen AI Use, Memory Migration, Nano Banana 2 Launches
Perplexity is pushing long-running multi-model agents, OpenAI is upgrading ChatGPT for better conversation and fewer hallucinations, Pew finds teens are mainstreaming AI in life, and other highlights.
Welcome to this week’s edition of AImpulse, a five point summary of the most significant advancements in the world of Artificial Intelligence.
Here’s the pulse on this week’s top stories:
Perplexity launches Perplexity Computer, a multi-model agent platform
What’s Happening
Perplexity introduced Perplexity Computer, a new orchestration system that dispatches work across 19 different AI models. It positions model flexibility as a core product feature, making it one of the first major platforms to build around multi-model autonomy.
Details
Users describe an outcome, and the system launches sub-agents that can browse the web, write code, connect to apps, and complete tasks autonomously.
Each task runs inside its own sandbox, with the platform mixing and coordinating competing models depending on the job.
Perplexity says these agents can stay active for extended periods, potentially for months at a time.
Pricing is usage-based, with Max subscribers receiving 10,000 monthly credits and the option to manually choose which model handles a task.
Why It Matters
Multi-model choice has gradually entered AI products, especially in creative tooling, but Perplexity is pushing that flexibility into the agent layer itself. If it works as advertised, it could mark an early shift toward more durable, sandboxed autonomous systems that aren’t locked to a single model provider.
OpenAI rolls out GPT-5.3 Instant as the new default ChatGPT model
What’s Happening
OpenAI released GPT-5.3 Instant, replacing the default ChatGPT model for all users with a version designed to improve conversational quality. The update focuses less on raw reasoning gains and more on fixing the tone, responsiveness, and usability issues that have frustrated users.
Details
OpenAI says the model was tuned to reduce unnecessary refusals and soften the “cringe” or overly preachy tone users had criticized.
The company claims hallucinations dropped by more than 25% on web search and nearly 20% on internal knowledge benchmarks.
GPT-5.3 Instant is also positioned as a stronger writing partner, with improvements in web answers and information presentation.
OpenAI hinted that a larger upgrade may be close, teasing on X that “5.4” is coming sooner than expected.
Why It Matters
This release shows OpenAI responding directly to a growing complaint that personality and tone matter almost as much as raw intelligence in everyday use. But while improving the user experience may help, it may not be enough on its own if broader trust and adoption issues continue to build.
Pew study finds teens are mainstreaming AI, especially for school
What’s Happening
Pew Research Center published a new study showing that U.S. teens are using AI at mainstream levels, especially for school-related tasks. While many see the technology as helpful, they also believe AI-assisted cheating is widespread.
Details
The survey of 1,458 U.S. teens and parents found common AI use cases include getting information, completing schoolwork, and entertainment.
Roughly 60% of respondents said AI-powered cheating is common among classmates, rising to 75% among teens who use AI themselves.
Teens were generally positive about AI’s role in their lives, citing convenience, learning, and efficiency as benefits.
40% of parents said they’ve never discussed AI with their child, highlighting a gap in awareness around chatbot use.
Why It Matters
Teens are becoming one of the first generations to integrate AI into everyday life while institutions are still figuring out how to respond. That creates real challenges for education, but also a major opportunity to shape healthier norms for how AI gets used.
Anthropic adds one-step memory imports from rival AI platforms
What’s Happening
Anthropic launched a new tool that lets users transfer saved preferences, instructions, and context from other AI platforms into Claude. The move comes as the company looks to capitalize on a wave of switching and new user growth.
Details
Users copy a provided prompt into their current chatbot, then paste the output into Claude to begin the transfer.
The import can bring over saved instructions, personal details, project context, and behavioral preferences from ChatGPT, Gemini, or Copilot.
Anthropic also expanded Claude’s memory feature to free users for the first time.
Claude Code received an auto-memory upgrade, allowing it to retain project context, debugging patterns, and workflow habits across sessions.
Why It Matters
Persistent memory is becoming one of the most important retention features in consumer AI. By making switching easier and opening memory to more users, Anthropic is turning migration friction into an acquisition strategy.
Google releases Nano Banana 2 with better quality at lower cost
What’s Happening
Google rolled out Nano Banana 2, an upgraded image generation model with better resolution, consistency, text rendering, and speed. The new version also arrives at roughly half the price of its predecessor and has climbed to the top of major text-to-image leaderboards.
Details
Nano Banana 2 ranked No. 1 for text-to-image on Artificial Analysis and LM Arena, while also placing No. 3 on image editing tasks.
The model supports up to 4K resolution across aspect ratios, while maintaining consistency across as many as five characters and 14 objects in a scene.
At around $0.07 per image, it is nearly 2x cheaper than both Nano Banana Pro and OpenAI’s GPT Image 1.5.
Nano Banana 2 is now the default image generator across Gemini and Google’s broader tool ecosystem, while Pro remains available for paid users.
Why It Matters
Google is narrowing the gap between premium-quality image generation and low-cost, high-speed output. That could accelerate a market shift where users no longer have to choose between affordability, speed, and frontier-level performance.