In partnership with

Welcome back!

OpenAI just hit the brakes, and the gas. After backlash from users over its newest GPT-5 model, the company reinstated GPT-4o for paid users, added manual model controls, and increased quotas. This launch showed that AI personality and usability may matter just as much as raw IQ power. Let’s unpack…

Today’s Summary:

• 🔁 OpenAI revamps GPT-5 rollout

• 📷 Google Imagen 4 Ultra tops image benchmarks

• 🦜 DeepMind’s Perch AI tracks wildlife

• 🧠 OpenAI eyes brain-interface startup Merge Labs

• 📚 Claude Sonnet expands to 1M tokens

• ⚡️ Google drops tiny Gemma 3 model

• 🛠️ 2 new tools

TOP STORY

OpenAI adds new GPT-5 controls and brings back GPT-4o for paid users

The Summary: OpenAI adjusted GPT-5 after a turbulent debut marked by user backlash and concerns over AI “attachment”. Following strong demand, CEO Sam Altman has restored access to the older GPT-4o for paying users, added manual controls to choose GPT-5’s “Auto,” “Fast,” and “Thinking” modes, and raised usage quotas.

Key details:

• New “Show additional models” toggle in settings lets paid users access previous models: GPT-4o, GPT-4.1 and o3

  

• API traffic doubled within 48 hours of GPT-5 launch, maxing out GPU capacity. OpenAI says it will double resources within five months

• Quota increases: Plus users now get 3,000 “Thinking” queries per week with 200k-token context before falling back to “Thinking mini”

• GPT-5-Thinking is stronger than o3 for reasoning and planning

• For code, Anthropic Opus 4.1 seems better, but it’s more expensive

• GPT-5 reduces “hallucinations”, making it more reliable overall

• Most casual chats feel similar between GPT-4o and GPT-5-Fast. The real gains appear when you enable GPT-5 Thinking mode

• Some users reported emotional attachment to GPT-4o

• Personality tuning is underway to make GPT-5 “warmer” without copying GPT-4o’s flattery. Long-term, users may be able to customize their own tone and style

Why it matters: GPT-5’s rocky landing is a stress test for AI interaction. This was the first major rollout where a “personality change” triggered backlash, not simply a matter of raw IQ points. OpenAI’s fast response shows that emotional tone, UX and reliability are an economic moat, and a liability if mishandled.

FROM OUR PARTNERS

Automate Your Work with AI Agents, No Code Needed

The Simplest Way To Create and Launch AI Agents

Imagine if ChatGPT and Zapier had a baby. That's Lindy.

With Lindy, you can build AI agents in minutes to automate workflows, save time, and grow your business. From inbound lead qualification to outbound sales outreach and web scraping agents, Lindy has hundreds of AI agents that are ready to work for you 24/7/365.

Stop doing repetitive tasks manually. Let Lindy's agents handle customer support, data entry, lead enrichment, appointment scheduling, and more while you focus on what matters most - growing your business.

Join thousands of businesses already saving hours every week with intelligent automation that actually works.

Get 400 Free Credits

GOOGLE

Google Imagen 4 ranks #1 image AI with 2K resolution and faster output

The Summary: Google has released its Imagen 4 text-to-image model through the Gemini API and Google AI Studio. The family now includes Imagen 4 Fast for fast image generation, alongside the flagship Imagen 4 and the detail-oriented Imagen 4 Ultra. Both Imagen 4 and Ultra support 2K resolution for sharper, more detailed visuals.

Key details:

• Imagen 4 was trained on a large, filtered dataset and initially released in May  2025 to a limited set of subscribers

• The Imagen 4 family, including Fast, Standard, and the new Ultra, is now generally available via the Gemini API and AI Studio

• Human evaluations from LMArena rank it as the #1 image model

• All generated images include SynthID watermarking

Why it matters: Imagen 4 Ultra’s crisp visuals, high resolution text rendering and 2K output make it ready for creative workflows with little need for post-editing. The Fast/Ultra combo gives teams iteration speed at low generation cost. With SynthID watermarking on every output, Google is laying down clear rules for provenance and traceability.

Try Imagen Ultra for free in AI Studio

FROM OUR PARTNERS

The AI Marketing Playbook Used by the Pros

Unlock the Power of AI With the Complete Marketing Automation Playbook

Discover how to scale smarter with AI-driven workflows that actually work. This playbook delivers:

• A detailed audit framework for your current marketing workflows

• Step-by-step guidance for choosing the right AI-powered automations

• Pro tips for improving personalization without losing the human touch

Built to help you automate the busy work and focus on work that actually makes an impact.

Steal the Playbook

GOOGLE

DeepMind Perch AI listens to the music of birds

The Summary: Google DeepMind’s Perch 2.0 is an open-source AI model that processes audio datasets from forests and oceans. Trained on over 1.5 million labeled recordings spanning birds, mammals, amphibians, insects, and human-made noise, it achieves top performance in detecting rare bird populations.

Key details:

• Training expanded from bird-only to nearly 15,000 animal and sound classes, including marine species

• Processes millions of hours of audio, identifying species, individuals, and even specific calls

• Uses “agile modeling” so scientists can build a new high-quality classifier in under an hour from a single sound example

• Works on consumer-grade laptops

Why it matters: Perch AI compresses what once took years of manual audio review into hours, letting conservation teams respond while endangered animal species are still recoverable. Its ability to detect rare bird calls opens new possibilities, from tracking birds recovery in treated mosquito zones to mapping reef soundscapes as indicators of marine health. Open-sourcing the model means any researcher can build accurate classifiers in under an hour at no cost, turning field recordings into actionable biodiversity maps.

Research paper

QUICK NEWS

Quick news

• Claude Sonnet now supports 1M tokens of context

• OpenAI plans to invest in new brain-computer interface startup

• Google releases Gemma 3 270M tiny model for hyper-efficient AI

TOOLS

🥇 New tools

• Kuse - Work with AI on a structured canvas

• Snowglobe - Simulate real users to test your AI before launch