Another month has passed, and the world of Artificial Intelligence (AI) has gotten more exciting, accessible, and deeply embedded in our daily lives. Nothing’s stopping the developments from happening at a revolutionary pace. From major tech giants to new players, everyone’s pushing the boundaries to create unmatched digital experiences.
So, without further ado, let’s explore the latest AI breakthroughs for June 2025.
Top 10 AI Model Releases You Should Not Miss
Here’s a look at the most groundbreaking AI model releases making waves across the industry.
1. OpenAI’s o3-pro
Topping the list in AI’s latest news is o3-pro, OpenAI’s newest brainchild. This reasoning model, touted as the company’s most capable model to date, is built on the foundation of OpenAI’s o3. It is designed to:
- Tackle problems with a meticulous, step-by-step approach, making it a game changer for fields such as math, physics, and coding.
- Search the web with a suite of tools capable of analyzing files, interpreting visual inputs, utilizing Python, personalizing its responses leveraging memory, and much more.
Poised to elevate the experience, o3-pro is now available for ChatGPT Pro and Team users, replacing the o1-pro model. Edu and Enterprise users will gain access later. This AI model is also live in OpenAI’s developer API. This opens the door to endless possibilities for developers and innovators alike.
Moreover, o3-pro has already demonstrated impressive prowess in AIME 2025, a benchmark test designed to evaluate a model’s math skills, outshining Google’s Gemini 2.5 Pro. It also beat Anthropic’s recent release, Claude Opus 4, on the challenging GPQA Diamond, a test that assesses PhD-level science knowledge.
2. Midjourney’s V1
If you are an artist looking to bring your images to life through video, Midjourney’s V1 is the game-changing AI video generation model for you. This much-awaited image-to-video model allows users to upload a still image or choose an image generated by Midjourney’s other models and transforms them into a set of videos with a duration of 4–5 seconds.
With the launch of V1, Midjourney has leaped into fierce competition with heavyweight contenders such as:
- Runway’s Gen 4
- Adobe’s Firefly
- OpenAI’s Sora
- Google’s Veo 3
However, what sets Midjourney apart is its distinctive AI image models catering to creative types—a stark contrast with competitors’ focus on developing controllable AI video models for use in commercial settings.
And the availability? Well, V1 is currently web-exclusive and primarily operates through Discord.
3. Microsoft’s Mu
Making waves in the tech world is Microsoft’s Mu, a new AI on-device Small Language Model (SLM). The tech giant has also released new features for Windows 11 in beta, including the new AI agents feature in Settings. The feature allows users to describe exactly what they want to do in the Settings menu and uses AI agents to either navigate to the right option or perform the action autonomously.
According to Microsoft, Mu is:
- Deployed entirely on-device in compatible Copilot+ PCs.
- Designed to run on the device’s Neural Processing Unit (NPU).
Mu is built on a transformer-based encoder-decoder architecture with an impressive 330 million token parameters, which makes it good for small-scale usage. In such an architecture, the encoder first converts the input into a comprehensible, fixed-length format, which the decoder then analyzes and generates precise outputs. With Mu, Microsoft is pushing the boundaries of refining the user experience.
4. Meta’s V-JEPA 2
Meta has just rolled out V-JEPA 2, a world model exclusively designed to enable AI agents to comprehend their surroundings. This model is built on the foundation of its predecessor, V-JEPA, which was trained on over a staggering 1 million hours of video. This training is supposed to help robots and AI agents operate in the physical world, predicting and understanding how concepts like gravity will affect sequential actions.
So, how does this new version fare? Meta claims that this enhanced model is an astounding 30 times faster than NVIDIA’s Cosmos model, which also tries to strengthen intelligence related to the physical world.
Nevertheless, here’s a point to note: Meta may evaluate its own models based on benchmarks that differ from those used by NVIDIA.
Also Read: AI Updates for May 2025: Revealing Latest Trends and Innovations
5. Google’s Gemini 2.5 Flash-Lite
In yet another exciting AI news update, Google unveiled its brand-new Gemini 2.5 Flash-Lite model, touted as the organization’s fastest and most cost-efficient AI powerhouse to date. Believed to perform better than its predecessor (2.0 Flash-Lite), this model excels in multiple areas, including:
- Coding
- Mathematics
- Science
- Reasoning
- Multimodal tasks
Users can now access the stable versions of Gemini 2.5 Pro and Gemini 2.5 Flash models, as Google has made the Gemini 2.5 family of AI models generally available. Google AI Pro and Ultra users can access the Gemini 2.5 Pro model without restrictions. And here’s a delightful surprise: The Pro model is also made available to users on the free tier of the Gemini platform; however, they have a daily limit.
Want to try this new model? It is currently available on Google AI Studio and Vertex AI, where you will find the stable versions of 2.5 Pro and Flash. Google is also weaving the capabilities of the 2.5 Flash-Lite and Flash models into its Search functionality.
6. Mistral AI’s Magistral
Say hello to Magistral, Mistral AI’s first family of reasoning models. This model adopts a stepwise approach for enhanced consistency and reliability when addressing problems related to topics such as math and physics—an approach used by other reasoning models like o3 and Gemini 2.5 Pro.
Users looking for reasoning AI models will be delighted to know that Magistral comes in two versions:
1. Magistral Small: This 24 billion parameter model can be downloaded from the AI dev platform Hugging Face with an Apache 2.0 license.
2. Magistral Medium: This advanced, more capable model is currently in preview on Mistral’s Le Chat chatbot platform, the company’s API, and third-party partner clouds.
7. Google’s Gemma 3n
The most recent artificial intelligence news comes from Google. In its blog post published on June 26, 2025, the tech giant announced the release of the full version of Gemma 3n, the latest open-source model in the Gemma 3 family of AI models.
This multimodal AI model packs quite a punch. It is:
- Engineered for on-device applications
- Designed with several new architecture-based improvements
- Capable of running locally on a mere 2 GB of RAM, making it even ideal for smartphones equipped with AI-enabled processing power
- Capable of supporting audio, video, image, and text inputs; outputs are limited to text only
- Designed with multilingual capabilities that support an astonishing 140 languages for texts and 35 languages for multimodal inputs
There’s more to know about Gemma 3n. Thanks to its mobile-first architecture built on the Matryoshka Transformer or MatFormer architecture, this Large Language Model (LLM) puts mobile capabilities front and center. Available in two versions, E2B and E4B, this model is tailored to meet varied needs.
Eager to explore Gemma 3n? The model is currently available for download through Google’s Hugging Face listing and Kaggle listing. Users can also dive right into action at the Google AI Studio, from where they can deploy Gemma models directly to Cloud Run.
8. Tencent’s HunyuanPortrait
Tencent’s HunyuanPortrait is designed to breathe life into ordinary, still portraits, transforming them into captivating animations. Powered by diffusion architecture, this LLM can whip up videos with realistic animation, capturing facial expressions and spatial movements, all based on a reference image and a guiding video. The ingenuity of this technology lies in its ability to ensure that the animated features synchronize perfectly with the original (reference) image.
Tencent’s post on X says that HunyuanPortrait is:
- Now open-source
- Available to the open community for downloading and experimenting from popular repositories like GitHub and Hugging Face listings
- Available for academic and research-based applications, but not meant for commercial usage just yet
From researchers to art enthusiasts, HunyuanPortrait offers a thrilling glimpse into the future of digital creativity.
9. Vercel’s v0
Vercel’s launch of the v0 AI model marks the first AI model to be developed by the company, specialized for web application development (both front-end and back-end). This model, available in beta, can be accessed through the company’s API as an AI Software Development Kit (SDK). It is also available via AI Playground.
In its X post, Vercel announced the release of its v0’s AI model—the same model that powers the Vibe coding platform v0. The company says that the model, officially named v0-1.0-md:
- Specializes in website development knowledge
- Integrates with OpenAI’s API
- Provides swift and relevant responses
- Supports both text and image inputs
10. Hugging Face’s SmolVLA
Hugging Face is all set to revolutionize robotic workflows and training-related tasks with SmolVLA, an open-source Vision-Language-Action (VLA) AI model. The company says that despite the rapid growth of AI technology, advancements in robotics still lag due to a shortage of high-quality and diverse data and LLMs that are designed for robotics workflows.
SmolVLA, an LLM, is designed to cement these gaps faced by the robotics research community. With 450 million parameters, this open-source robotics-focused model is trained on an open dataset provided by the LeRobot community. The company claims that this AI model is compact yet powerful and is capable of outperforming significantly larger models. It can run seamlessly on a desktop computer with a single compatible GPU or even newer sleek MacBook devices.
Robotic enthusiasts can download and experiment with it right away. So, gear up to explore the future of AI-driven robotics with SmolVLA!
Now that we have explored the top ten new launches for June, let’s move on to the latest updates on the existing AI models.
Also Read: AI Updates for April 2025: The Latest From the Tech World
Important Latest AI Updates for June 2025
Here’s a roundup of the most significant AI updates for June 2025, showcasing key improvements to some of the most advanced models in use today.
Updates to the Gemini 2.5 Pro AI Model
Google has just rolled out an updated preview of its Gemini 2 5 Pro model. Users can now access this enhanced version through AI Studio, Vertex AI, and the Gemini app, with general availability (expected) in another couple of weeks.
In a recent blog post, Google explains that Gemini 2.5 Pro is not just a minor upgrade but a coding powerhouse, particularly excelling in breezing through some of the toughest coding benchmarks available. It also demonstrates impressive abilities in math, science, and reasoning, essentially proving itself a top-tier contender in AI capabilities.
DeepSeek’s Updates on R1 AI Model
DeepSeek’s updated R1 reasoning model, described as a “minor trial upgrade,” runs on a single GPU. This updated model comes alongside a more compact, distilled version, known as DeepSeek-R1-0528-Qwen3-8B, which is already claimed to outperform its similar-sized counterparts on select benchmarks. The smaller updated R1 performs better than Google’s Gemini 2.5 Flash in the AIME 2025 competition and also nearly matches Microsoft’s Phi‑4‑reasoning‑plus model on HMMT.
Mistral’s Updates on Its Open-Source Small Model From 3.1 to 3.2
Mistral is making strides by updating its 24B parameter Mistral Small model from version 3.1 to 3.2-24B-Instruct-2506. This upgrade enhances the capabilities of Mistral Small 3.1, aiming to refine specific behaviors, such as instruction adherence, output stability, and function calling robustness.
Instruction-following benchmarks show a slight yet noticeable improvement. The internal accuracy of Mistral has increased from 82.75% in Small 3.1 to 84.78% in Small 3.2. This demonstrates a significant leap in instruction-following performance.
Exciting AI Latest Updates in June
Let’s now explore the most exciting June AI updates that introduce powerful features, creative tools, and enhanced experiences across top AI platforms.
Character.AI’s Video Generation and Social Feeds
Character.AI has revealed a treasure trove of multimedia features in a recent blog post. These features include:
- AvatarFX, a video-generation model set to elevate user interaction
- Scenes and Streams to craft captivating videos featuring their beloved characters and then share them on social media feeds
Anthropic’s Claude Updates
-
Voice Mode for Claude
Anthropic’s Claude chatbot apps now have their own voice mode, which is currently available in beta. The company’s post on X and the updated documentation on their website say that Claude mobile app users can have natural, complete spoken conversations with the AI companion in English (only) across all plans.
-
Creativity With Claude
Anthropic has yet another fascinating update for Claude. This new AI feature allows users to dive into a world of app creation, building interactive AI-powered applications directly through the chatbot. Users can also host and share these apps within the platform. Interestingly, when others use these apps, the maker need not worry about incurring API usage charges, which will be charged against the user’s subscriptions.
Perplexity’s Leap Into Video Generation
Perplexity’s AI chatbot on X, Ask Perplexity, now comes with an innovative video generation capability. Users can now create 8-second videos simply by tagging the bot in posts or first-level comments. The bot uses text prompts and text-plus-image inputs to bring these videos to life with native audio and even multi-speaker dialogues. Talk about elevating your social media presence!
OpenAI’s ChatGPT Updates
-
Memory Improvements for Free Users
OpenAI is rolling out exciting memory improvements that allow all ChatGPT users to enjoy personalized responses based on their previous conversations. Free users will receive a slightly simplified version of this feature, alongside Codex, its coding-focused AI agent, to gain full access to the internet that can be activated during the setup phase.
-
Web Search Tool Upgrades
OpenAI has quietly turbocharged its web search tool within ChatGPT to:
- Enhance the quality of web search-based responses.
- Boost the assistant’s ability to follow instructions.
In addition to these, this tool can also handle longer and more complex queries. Also, a notable enhancement has been made to image-based web searching. To top it off, OpenAI has expanded ChatGPT’s projects feature and made the Canvas tool downloadable, thus providing more flexibility.
ElevenLabs Expands Text-to-Speech Model
ElevenLabs has unveiled the language expansion of the Eleven v3 text-to-speech (TTS) model. The company announced in a post that its latest AI TTS model now supports 41 new languages, bringing the total language count to over 70—a leap forward for users looking to engage in lifelike conversations.
Text-to-Video Service With Manus
Manus’s post on X unveils a text-to-video generation feature that requires just a few minutes to transform a simple text command into a structured video narrative. If you are a paid subscriber, Manus gives you early access to the tool before others can use it for free of cost.
Google Supercharges Gemini Code Assist
In an exciting update, Google has breathed new life into its Gemini Code Assist platform’s capabilities with the Gemini 2.5 Pro AI model. The platform is free of charge, and users can also experiment with its enterprise version. Upgrades also include a better chat function and new personalization features.
Also Read: AI Insights for March 2025: Unveiling the Latest Innovations
AI New Updates: Key AI Tools and Agents for June 2025
Let’s explore the AI latest updates for June 2025, featuring new tools and agents from top tech platforms.
Google’s Latest Tools and Agents
- Google is enhancing its AI Mode with features that facilitate back-and-forth voice conversations. Its experimental Search feature allows users to ask complex, multi-part questions. Additionally, it can now create interactive graphs and charts, making data visualization more accessible for understanding data patterns when analyzing financial trends.
- Google has announced the expansion of Gemini in Google Docs to Android devices, which was previously exclusive to paid Workspace users on the desktop version. Users can now leverage Gemini’s capabilities while using Docs on Android devices.
- Google has also unveiled Gemini CLI, a cutting-edge open-source AI command-line interface tool. This innovative AI agent aids developers with daily coding tasks and can also autonomously execute actions. It is available for free in preview on GitHub, and it is a game-changer for coding efficiency.
Microsoft’s Latest Tools and Agents
- Microsoft Bing is rolling out the Bing Video Creator to its app, powered by OpenAI’s Sora model, enabling users to generate videos from text prompts. However, access to Sora’s video generation remains exclusive to paying customers.
- MS Paint, Snipping Tool, and Notepad will have exciting new AI features as a part of an update for Windows Insiders in the Canary and Dev channels, exclusively for Copilot+ PC users. The update includes a sticker generator and element selection for MS Paint, easier screenshot capture in Snipping Tool, and an AI-driven Write feature in Notepad to expand or generate text.
Canva’s Video Clip Feature
Canva has unveiled a dynamic new addition to its AI suite: “Create a Video Clip,” leveraging Google’s Veo 3 video-generation AI model. This text-to-video tool offers cinematic-quality video rendering and native audio generation. This tool is available for paid subscribers, Canva non-profit users, and users on the Leonardo.ai platform.
Mistral’s Mistral Code
Mistral has introduced Mistral Code, an AI-powered coding assistant that combines its models, an “in-IDE” assistant, local deployment options, and enterprise tools into one comprehensive package derived from the open-source project Continue.
Yahoo’s New Mail App Features
Yahoo is adding a new gamified “Catch Up” feature to its mail services aimed at simplification. This feature provides AI-powered summaries and email previews, giving users the option to “delete” or “keep in inbox.”
Adobe’s AI-Powered Image and Video Tools
Adobe released its Firefly platform as a mobile app for Android and iOS users, featuring native AI models and also third-party models for image generation, video generation, and photo editing. Free users get limited credits, and paid users receive plan-based credits.
Meta’s Generative AI-Powered Video Editing Tool
This new AI-powered video editing feature has preset templates that allow users to edit and enhance short-form videos by changing the background, adding new effects to the subjects, and even changing the clothing of people. This template-style generative AI-powered tool is available for free for a limited time and can be accessed on the Meta AI app, Meta.AI website, and Edits app.
ChatGPT’s Upgrade on WhatsApp
For users accessing ChatGPT via WhatsApp, it has now been upgraded to generate images based on user prompts, answer text-based questions, create content, respond to voice notes in text only (two-way conversation is not supported), and analyze images.
Upcoming AI Updates to Watch Out for
Here are some exciting AI news items to look out for in the forthcoming weeks and months:
- xAI’s Grok will soon come equipped with a file editor feature that might support various file types, including spreadsheets.
- Google’s AI chatbot Gemini will soon get a quality-of-life improvement that enables users to select specific portions of Gemini’s responses by directly long-pressing and dragging and then quickly share them with other apps.
- The Gemini AI assistant will work with several apps on Android devices, regardless of whether the feature is enabled or not. These apps include Messages, WhatsApp, Utilities, and the Phone app.
- Later this year, Google plans to launch SignGemma, a new open-source AI model that can translate sign language into spoken text. It will help individuals with speech and hearing disabilities communicate with those who don’t know sign language.
In conclusion, AI’s new updates from all over the world give us a promising glimpse into a future where technology and our routine lives merge as one. With platforms like Maayu.ai leading the way, we are witnessing simplified complexity and user empowerment.
As we look ahead to July, there is palpable excitement for the innovations yet to come, confirming that the future of intelligent technology is well within our reach.