Microsoft launches Phi 4 reasoning AI models to compete with DeepSeek and OpenAI
Microsoft has launched a series of AI reasoning models - Phi 4 mini reasoning, Phi 4 reasoning and Phi 4 reasoning plus, as a part of its small language model family. The open-weight models combine reinforcement learning with supervised fine-tuning to solve complex problems in maths, science, coding and logic-based tasks.
The most capable one of them - Phi 4 reasoning is a 14-billion parameter model that was trained on curated and high-quality web data from OpenAI's o3-mini. A blog posted by the company said that the AI model performs the best at math, science and coding applications.
Meanwhile the Phi 4 mini reasoning model has 3.8 billion parameters and was trained on around a million synthetic math problems generated by DeepSeek's R1 model. The AI model is built for teaching purposes and can be embedded into devices.
The Phi 4 reasoning plus on the other hand outperforms DeepSeek R1 on multiple benchmarks even though the latter is much bigger in size. The model is also comparable to OpenAI's o3-mini on a math test benchmark.
'Phi's evolution over the last year has continually pushed this envelope of quality vs. size, expanding the family with new features to address diverse needs. Across the scale of Windows 11 devices, these models are available to run locally on CPUs and GPUs,' Microsoft stated.
The mini reasoning models are all available now over Hugging Face.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
India Today
2 hours ago
- India Today
Anthropic working on building AI tools exclusively for US military and intelligence operations
Artificial Intelligence (AI) company Anthropic has announced that it is building custom AI tools specifically for the US military and intelligence community. These tools, under the name 'Claude Gov', are already being used by some of the top US national security agencies. Anthropic explains in its official blog post that Claude Gov models are designed to assist with a wide range of tasks, including intelligence analysis, threat detection, strategic planning, and operational support. According to Anthropic, these models have been developed based on direct input from national security agencies and are tailored to meet the specific needs of classified introducing a custom set of Claude Gov models built exclusively for US national security customers,' the company said. 'Access to these models is limited to those who operate in such classified environments.'Anthropic claims that Claude Gov has undergone the same safety checks as its regular AI models but has added capabilities. These include better handling of classified materials, improved understanding of intelligence and defence-related documents, stronger language and dialect skills critical to global operations, and deeper insights into cybersecurity data. While the company has not disclosed which agencies are currently using Claude Gov, it stressed that all deployments are within highly classified environments, and the models are strictly limited to national security use. Anthropic also reiterated its 'unwavering commitment to safety and responsible AI development.'Anthropic's move highlights a growing trend of tech companies building advanced AI tools for defence. advertisementEarlier this year, OpenAI introduced ChatGPT Gov, a tailored version of ChatGPT that was built exclusively for the US government. ChatGPT Gov tools run within Microsoft's Azure cloud, giving agencies full control over how it's deployed and managed. The Gov model shares many features with ChatGPT Enterprise, but it places added emphasis on meeting government standards for data privacy, oversight, and responsible AI usage. Besides Anthropic and OpenAI, Meta is also working with the US government to offer its tech for military month, Meta CEO Mark Zuckerberg revealed a partnership with Anduril Industries, founded by Oculus creator Palmer Luckey, to develop augmented and virtual reality gear for the US military. The two companies are working on a project called EagleEye, which aims to create a full ecosystem of wearable tech including helmets and smart glasses that give soldiers better battlefield awareness. Anduril has said these wearable systems will allow soldiers to control autonomous drones and robots using intuitive, AR-powered interfaces.'Meta has spent the last decade building AI and AR to enable the computing platform of the future,' Zuckerberg said. 'We're proud to partner with Anduril to help bring these technologies to the American service members that protect our interests at home and abroad.'Together, these developments point to a larger shift in the US defence industry, where traditional military tools are being paired with advanced AI and wearable tech.
Hindustan Times
2 hours ago
- Hindustan Times
OpenAI eyes collaboration with India for data centres
OpenAI, which developed the artificial intelligence bot called ChatGPT, is in talks to partner with India under its 'OpenAI for Countries' initiative, a person familiar with the matter has said. 'OpenAI for Countries' is a new global initiative by the creator of ChatGPT to help interested national governments with efforts like building out data center capacity in coordination with the US government. OpenAI's chief strategy officer Jason Kwon, who is currently in Delhi, has been touring across Asia Pacific, covering countries like Japan, South Korea, Australia, India and Singapore, meeting governments and potential private-sector partners to discuss opportunities within the 'OpenAI for Countries' initiative. The 'for Countries' initiative works in a two-way partnership between the US and another country. In its pilot project announced with UAE last month, the middle=east country will house a 1-gigawatt AI computing cluster called Stargate in Abu Dhabi, while also investing in similar AI infrastructure in the US as part of the US-UAE AI Acceleration Partnership. While details of the India-US partnership under the 'OpenAI for Countries' initiative remain undisclosed, the company's goal is to pursue 10 projects with individual countries or regions as the first phase of this initiative, and expand from there, said a company blog post. In a step towards 'cultivating AI talent' in India, OpenAI, in partnership with MeitY's IndiaAI Mission, launched 'OpenAI Academy' to give India's developer community and network of startups and innovators access to AI education and tools. The two also signed a memorandum of understanding, making it OpenAI's first international launch of its education platform. The MoU outlines seven joint initiatives, including OpenAI contributing educational content to IndiaAI Mission's FutureSkills platform to support digital skilling. The educational content will also be hosted on the iGOT Karmayogi platform to train government officials, initially in English and Hindi, and later in four regional languages. Additionally, OpenAI will conduct regular webinars and in-person workshops in six cities, and provide up to $100,000 in API credits to 50 IndiaAI-approved fellows or startups. The partnership also includes a series of hackathons across seven states, expected to reach around 25,000 students, with winners receiving invitations to attend OpenAI Dev Day events internationally. 'We believe the latest AI frameworks and tools must be made widely accessible so that our startups, developers, and researchers can build faster, better, and at scale,' IT minister Ashwini Vaishnaw said in a video message at the event.
Hindustan Times
3 hours ago
- Hindustan Times
Microsoft tells Windows 10 users: Upgrade to Windows 11 or risk your security
Come October 2025, Windows 10 would reach the end of support from Microsoft. What does this mean? The operating system will no longer receive free security updates, new features, or technical support from Microsoft. Your PC will keep running, but without updates, it's wide open to hackers, malware, and security threats. As Microsoft tries to onboard more Windows 10 users to Windows 11, it's trying out novel ways and campaigns to achieve its goals. A new advertisement on the official Windows YouTube channel is urging people to be on the "right side of risk" by upgrading to Windows 11. While it's true that Windows 11 would be safer to use in the long run due to support from Microsoft, this ad is still getting a lot of attention. Why? Simply because Microsoft is trying to use the end of support for Windows 10 as a way to sell its current software for PCs. It's noteworthy that Windows 11 is not new. In fact, it's been around for almost 5 years. Windows 10, on the other hand, has been around for almost 10 years now and was launched on July 29, 2015. Microsoft's strict Trusted Platform Module (TPM) requirements have also drawn ire from users, with many saying that TPM 2.0 should not be necessary to run Windows 11. Before we talk about how this will negatively impact Windows 10 users, let's explore what TPM means. In short, it's a dedicated chip that is designed to give "hardware-level security services for your device." This means its job is safeguard your personal information and credentials from unauthorised users. Microsoft has made it clear that TPM 2.0 is non-negotiable requirement to upgrade to Windows 11. The problem with this is that many PCs and Windows laptops are now unable to move from Windows 10 because Windows 11 has specific hardware requirements. Essentially, they're now stuck on an older OS that may not provide enough security in the long run. Sure, Microsoft has announced an extended support programme for Windows 10, but it's definitely not a cheap option. Pricing for the same begins at $61 (over ₹5,000) for the first year, $122 (over ₹10,000) for the second year, and $244 (over ₹20,000) for the third year per device. That doesn't sound very customer friendly on paper, but this is how Microsoft is pushing all Windows users to upgrade to Windows 11. Will this strategy work? Only time will tell.
