Summary of the key AI information you asked about from this week (around late April 2025), based on recent online reports:
- GPT-4o Upgrade: OpenAI rolled out an update to GPT-4o, improving its intelligence, personality, conciseness, and ability to follow instructions and handle coding tasks. For ChatGPT Plus subscribers, the hourly usage limits for GPT-4o and related models were doubled. OpenAI also confirmed GPT-4 will be removed from ChatGPT on April 30th, making GPT-4o the default for all users.
- OpenAI gpt-image-1: This is the new image generation model powering ChatGPT's image creation feature. OpenAI has now made gpt-image-1 available via API, allowing developers to integrate it into other applications. Major platforms like Adobe (Firefly, Express), Figma, Canva, GoDaddy, and HubSpot are integrating or exploring integration of this model.
- ChatGPT Deep Research Mini: OpenAI expanded access to its "Deep Research" feature by introducing a new lightweight version powered by the o4-mini model. This version is rolling out to all users, including the free tier, offering faster research capabilities that are less resource-intensive, although responses may be more concise.
- GPT o3 / o4 mini / Deep research Usage Adjustments: With the rollout of the lightweight Deep Research (o4-mini), OpenAI introduced monthly query limits: 5 for free users, 25 for Plus/Team/Enterprise/Edu, and 250 for Pro users. If the limit for the full version (using the o3 model) is reached, users automatically switch to the lightweight version. Separately, hourly usage limits for GPT-4o were doubled for Plus subscribers.
- Grok Vision Feature: Elon Musk's xAI introduced "Grok Vision," enabling the Grok chatbot to interpret and answer questions about real-world objects, signs, or documents using a smartphone's camera in real-time. It launched initially on the Grok iOS app.
- Genspark AI Slides: Genspark AI added a feature to generate presentation slides. Users can upload documents (like Word, PDF) or other inputs, and the AI creates slides with research-backed content, charts, and visuals.
- Perplexity Voice Assistant: Perplexity launched its AI voice assistant for iPhones (it was previously available on Android). It allows users to perform tasks like playing music (Apple Music), managing calendars (Apple Calendar), sending emails (Apple Mail), and getting directions (Apple Maps) using voice commands. Support for third-party apps like Gmail and Google Calendar is planned.
- Perplexity New Features/Models (GPT Image Gen / Grok 3 / o4 mini): The main Perplexity news this week was the launch of its Voice Assistant on iOS. While Perplexity leverages various AI models, the search results didn't specifically confirm the addition of GPT image generation, Grok models, or o4-mini as distinct feature announcements this week.
- Dreamina AI's Top AI Image: Dreamina's AI image generator and logo generator were highlighted in articles discussing AI's role in creating visuals for television branding and graphics and for designing unique digital collectible stickers.
- Tavus SoTA lipsync model: AI company Tavus launched "Hummingbird-0," described as a new state-of-the-art (SOTA) lip-sync model achieving high scores for realism, accuracy, and maintaining the speaker's identity.
- Dia SoTA speech AI model: A startup named Nari Labs released Dia, an open-source 1.6 billion parameter text-to-speech (TTS) model. It's designed to generate natural-sounding conversational speech directly from text, including emotional tones and non-verbal cues like (laugh) or (cough). It's positioned as a competitor to offerings from ElevenLabs and others.