March 2025 AI Platform Feature Announcements

So much happened in March that, in addition to our newsletter, we decided to summarise the developments within the leading platforms. Here goes (detail first and then a summary)...

ChatGPT (OpenAI)

• Built-in Image Generation: OpenAI’s ChatGPT (Plus tier) gained the ability to generate images from text prompts in late March 2025 (Internet Reacts To ChatGPT New AI Image Feature - Pubity). This update introduced a “natively multimodal” model capable of producing high-quality, stylistically diverse images (users even turned CEO Sam Altman into a Studio Ghibli-style cartoon) (Internet Reacts To ChatGPT New AI Image Feature - Pubity). See down below for an example of what ChatGPT can do now (and potentially save some $$ on that Midjourney subscription!).

• Advanced Audio Models: OpenAI launched new voice features for ChatGPT, including two state-of-the-art speech-to-text models (which outperform the Whisper system) and a text-to-speech model that follows style instructions (OpenAI, Claude, Gemini all copy each other). They also added audio integration in the ChatGPT Agents SDK, allowing developers to create voice-enabled AI agents (OpenAI, Claude, Gemini all copy each other).

Gemini (Google)

• “Canvas” Collaborative Workspace: Google’s Gemini assistant introduced Canvas, an interactive workspace for real-time collaboration (New Gemini features: Canvas and Audio Overview). This lets users draft, edit, and share documents or code with Gemini’s help, enabling easier brainstorming, coding, and content creation directly in the chat interface.

• “Audio Overview” Summaries: Gemini also added an Audio Overview feature that converts documents into podcast-style audio discussions (New Gemini features: Canvas and Audio Overview). In practice, the AI can summarise a file’s key points and present them as an engaging audio narration by virtual “hosts,” helping users digest information in a spoken format.

Claude (Anthropic)

• Web Browsing Capability: Anthropic’s Claude AI assistant can now “surf the web” to fetch up-to-date information (OpenAI, Claude, Gemini all copy each other). This March 2025 update allows Claude to perform live web searches (similar to ChatGPT’s browsing or tools like Perplexity), returning answers with cited sources for verification (OpenAI, Claude, Gemini all copy each other). The feature debuted for paid US users (with a wider rollout expected) and can be enabled via a settings toggle (OpenAI, Claude, Gemini all copy each other).

Grok (xAI)

• Image Editing Feature: Elon Musk’s xAI added a new image-editing tool to Grok in March 2025. Users can upload an image and describe desired edits; Grok will then generate a modified version of the image reflecting the request (Grok (chatbot) - Wikipedia). This enables on-the-fly image manipulations (e.g. changing background or style) directly through the chatbot.

• “DeeperSearch” Engine: Alongside image editing, xAI rolled out DeeperSearch – an enhanced version of Grok’s earlier DeepSearch feature (Grok (chatbot) - Wikipedia). DeeperSearch provides more extensive web research and reasoning capabilities, allowing Grok to conduct deeper multi-step searches and articulate its thought process more clearly when answering complex queries.

DeepSeek

• DeepSeek V3-0324 Model Upgrade: Chinese AI firm DeepSeek released an upgraded version of its V3 model in March 2025 (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET) (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET). The new V3-0324 model delivers a “major boost in reasoning performance” and stronger coding skills (especially for front-end web development), along with “smarter tool-use capabilities,” according to the company (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET). It builds on DeepSeek’s previous open-model approach, with the updated weights made publicly available under an MIT license (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET) (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET).

• Enhanced Functions and Accuracy: DeepSeek V3-0324 introduced improved function-calling abilities (DeepSeek's latest AI release takes another swipe at industry giants | Capacity Media) – meaning the model can more reliably fetch external data or use tools via API calls, a weak point in the prior version (DeepSeek's latest AI release takes another swipe at industry giants | Capacity Media). It also achieved better performance on benchmarks (for example, significantly higher scores on a challenging math exam) (DeepSeek's V3 AI model gets a major upgrade - here's what's new | ZDNET), indicating advancements in medium-to-long-form answers and factual accuracy.

Meta AI (Meta Platforms)

• European Rollout and Multilingual Support: Meta announced a major expansion of its Meta AI assistant across Europe in March 2025 (Europe, Meet Your Newest Assistant: Meta AI | Meta). The AI, which had over 700 million monthly users globally by this time (Europe, Meet Your Newest Assistant: Meta AI | Meta), began rolling out to 41 European countries (including all EU nations and several territories) for the first time (Europe, Meet Your Newest Assistant: Meta AI | Meta). Meta AI’s chat function initially supports six European languages, marking the first step toward parity with the English-language version in the US (Europe, Meet Your Newest Assistant: Meta AI | Meta).

• Group Chat Integration: Meta AI gained new integration within group chats on the company’s messaging platforms. Users can now summon the assistant in WhatsApp group conversations (with rollout to Messenger and Instagram Direct chats to follow), allowing the AI to help multiple people in a chat with answers or suggestions (Europe, Meet Your Newest Assistant: Meta AI | Meta). This on-call group assistant feature enhances social use cases – for example, helping plan events or answer questions collaboratively in-chat.

TL;DR

All major AI platforms rolled out noteworthy new features, often echoing each other’s capabilities. OpenAI’s ChatGPT introduced new built-in image generation as well as powerful new voice (speech-to-text and text-to-speech) models, making the AI more multimodal than ever.

Google’s Gemini added a shared Canvas workspace for live document editing and coding, along with Audio Overview to turn files into audio summaries.

Anthropic’s Claude finally gained internet access, able to browse the web and cite sources in its answers.

Elon Musk’s Grok (xAI) similarly expanded with an image editor and a smarter DeeperSearch tool for reasoning-intensive queries.

DeepSeek, a rising competitor from China, upgraded its model (V3-0324) to boost reasoning, coding, and tool use, continuing its trend of open-weight releases.

Meanwhile, Meta’s AI assistant vastly expanded its reach – launching across Europe with multi-language support and integrating into group chats – underlining a race where each platform is rapidly adopting multimodal, collaborative, and real-time capabilities to one-up each other.

Generated with ChatGPT's new image generation tool