Q1 Roundup 2026 - Global AI Model Releases and Capabilities
I’ve never been able to last in a job or role for more than 2 years. Never. I always get bored by the end of the 2nd year if I last that long. I was constantly seeking an interesting challenge and work that left me excited, inspired and challenged.
Guess what? I’ve been in the field of AI for close to 7 years now. Not 2 years - 7 years! And there’s no end in sight. You know why? It’s a fast-changing challenging and super interesting field to work in.
This newsletter update is not about luring you into the field of AI. It’s an update on some of the latest AI news that occurred in AI in the first quarter of this year which you should be aware of.
We’ll look at the latest model releases and capabilities that were introduced in Q1 across the US, China and the rest of the world. We’ll also briefly analyse the current state of play with the impending AI bubble.
The AI wars will last for a very long time, probably forever. China and the US are still head to head in the AI war with the leading US providers constantly releasing new and improved models. The number of AI model releases are quite significant. Here are a few notable launches:
OpenAI: Open AI released GPT-5.4, GPT-5.3-Codex, GPT-5.4 mini + nano. GPT-5.4, considered as OpenAI’s latest flagship release combines reasoning, coding and agentic workflows in one frontier model. All other models are focused on coding, cybersecurity, smaller and faster versions optimised for coding, multimodal reasoning and high-volume API use. Open AI retired 3 models in Q1: GPT-4o, GPT-4.1 and GPT5.1.
Google Deepmind: Google’s approach has been more deliberate and intentional, with the focus on quality rather than quantity. Gemini 3 and Gemini 3.1 Pro were released in Q1, featuring a “Deep Think” function and smaller multimodal models for robotics vision, and vision and language tasks were released in early April.
Anthropic: Anthropic released Claude 4.6 Opus known for its nuanced, strategic, and deep analytical capabilities. Claude Sonnet 4.6 was launched in February, built for high-volume, enterprise tasks. Claude Sonnet specialises in agentic coding, advanced reasoning and computer use capabilities. Claude Opus 3 and Claude 3 Haiku have served their time and have both been retired.
Meta: Early this month, Meta released Muse Spark, a multimodal AI reasoning model with advanced agent capabilities. Meta is heavily focused on developing superintelligent AI models which it positions Muse Spark as the first step towards this feat.
xAI: We can’t fail to mention Elon Musk’s Grok which had some updates this year to improve its reasoning, hallucinations, X data integration and image / video capabilities. Grok 4.20 was released in February, built for speed and agentic tooling with 4 specialised AI agents running in parallel.
Microsoft: Didn’t quite make it into Q1, but at the early start of Q2 Microsoft announced the release of three foundational models known as MAI models. These models work across speech-to-text, text-to-speech and image / video.
Deepseek: DeepSeek V4 was released a couple of days ago. DeepSeek V4 is specialised for advanced coding, complex reasoning and high-efficiency / low cost operations, with the ability to run locally.
Alibaba: Alibaba released Qwen 3.5 just before the Chinese New Year. Costing a fraction of the US frontier models at $0.40 per million tokens (compared to GPT-5.4 which costs $2.50 per million tokens), it supports 201 languages, and handles multimodal inputs. Unlike the closed-source AI models from the US, Qwen and most of the AI models developed in China are open-source, making it easier to access and work with. Concerns around data and privacy still exist when working with models from China.
ByteDance: ByteDance, the parent company of TikTok based in China, released its GenAI chatbot in February, and is on par with ChatGPT and Gemini according to Reuters, carrying out complex reasoning and mult-step execution tasks.
Zhipu: Zhipu, a Chinese AI Unicorn focused on agentic engineering and multimodal models released its open-source GLM-5 model a few days before the Chinese New year. The model is engineered for agentic intelligence, advanced multi-step reasoning, and frontier-level performance in coding, creative writing and problem solving.
It may seem like Apple missed the invitation to the party, but they decided to partner with Google (dropping the previous partnership with OpenAI) and entered a multi-year collaboration where their models will be based on Google's Gemini models and Cloud Technology.
And while Donald Trump and Xi Jinping are still going head to head in the AI race, the rest of the world haven’t been sleeping.
Mistral, a start-up based in France is still leading in Europe and released a series of models this year including Mistral Small 4 and Mistral Forge.
The UAE also released a series of models including Falcon H1 Arabic - a large Arabic-language model. The UAE is focused on optimising models for Arabic languages.
In Canada, Cohere released a few models on reasoning and vision, and open-weight multilingual models supporting over 70 languages.
South Korea launched its AI Basic Act in January, one of the first countries to have a complete law on AI, and has a $735B sovereign AI initiative underway - another prediction I made last year on an increase in AI sovereignty.
The impending “AI Bubble Burst” is still looming with notable changes to GPT-4’s intelligence and accuracy, and price cuts by Open AI and Anthropic.
Why is any of this important? We’re clearly in a different age and era. And even if you don’t care too much about AI, it’s no longer just a technology, it’s a system and an integral part of our everyday lives.
While all the mentioned companies are the leading AI providers in the world, there are so many start-ups and third party applications that have spun off from these that are quite useful.
For example, my latest AI executive assistant, built by Lindy AI is a fine mix of Claude, Gemini and ChatGPT among other models and is able to work with different models and agents providing top-notch results.
It’s important to note that while the release of models and increased capabilities are not letting up, the speed of responsible AI and AI safety practices to fix, reduce and mitigate the damaging issues of AI with regards to improved accuracies, reduced hallucinations and sycophancy, bias, privacy and security concerns, data labour and colonialism, and so on are still very much behind. Leading AI providers need to ensure they prioritise the safety of their users as they race to outbid themselves in the never-ending race.
Policy makers and regulation also need to speed up and either catch up with the pace of development (a bit impossible) or create policies to slow down the pace or pause the massive scale of highly-risky models while mitigations are developed.
As I also predicted at the end of 2025, agentic AI is now a mainstream AI technology that is being integrated into workflows, enterprises and locally.
I anticipate the next 3 quarters will see an increase in agentic capabilities while the race continues. And while all of these take place, you need to make sure you’re building your AI literacy skills on a continuous basis and only work with the AI tools that give good results.