AI2 Tulu-3 surpasses DeepSeek V3 💡, Mistral Small 3 🔥, OpenAI o3-mini ⚡

In partnership with

Your daily AI dose

Mindstream is your one-stop shop for all things AI.

How good are we? Well, we become only the second ever newsletter (after the Hustle) to be acquired by HubSpot. Our small team of writers works hard to put out the most enjoyable and informative newsletter on AI around.

It’s completely free, and you’ll get a bunch of free AI resources when you subscribe.

Click here to subscribe

💬 OpenAI Developments

OpenAI o3-mini

OpenAI has introduced o3-mini, a cost-effective reasoning model optimized for STEM fields, offering enhanced performance in science, math, and coding tasks. Available through ChatGPT and the API, o3-mini supports features like function calling and structured outputs, and allows developers to select from low, medium, or high reasoning effort to balance speed and accuracy according to their needs.

OpenAI in Talks for Huge Investment Round Valuing It at Up to $300 Billion

OpenAI is reportedly in discussions to raise a substantial investment round, led by SoftBank, that could value the company at up to $340 billion. This follows a recent funding round in October 2024, where OpenAI nearly doubled its valuation to $157 billion after raising $6.6 billion from investors including Microsoft and Nvidia.

🚀 Tech Industry Moves

NVIDIA has announced the availability of DeepSeek-R1, a 671-billion-parameter reasoning model, as a NIM microservice preview on build.nvidia.com. This model excels in tasks requiring logical inference, reasoning, math, coding, and language understanding, delivering up to 3,872 tokens per second on a single NVIDIA HGX H200 system.

Ai2 says its new AI model beats one of DeepSeek’s best. On January 30, 2025, the Allen Institute for AI (Ai2) announced a new AI model that reportedly outperforms DeepSeek V3, a leading model from the Chinese AI company DeepSeek. This development highlights the rapid advancements in AI research and the increasing competition among organizations to develop more capable models.

Mistral AI has released Mistral Small 3, a 24B-parameter, latency-optimized open-source model under the Apache 2.0 license. It rivals much larger models like Llama 3.3 70B while being 3x faster on the same hardware. Designed for fast-response tasks, low-latency function calling, and local deployment, it performs well across code, math, and general instruction benchmarks. Mistral Small 3 is lightweight enough to run on a single RTX 4090 or a MacBook with 32GB RAM and is available on multiple platforms, including Hugging Face, Ollama, and Together AI. Mistral AI reaffirms its commitment to open-source and plans to release more models with enhanced reasoning soon.

DeepSeek AI blocked by Italian authorities as others member states open probes. The Italian Data Protection Authority has blocked DeepSeek AI due to concerns over its data privacy practices, specifically its handling of Italians' personal information. Other EU member states have also launched investigations into the company's compliance with GDPR, reflecting growing scrutiny over AI models' data usage across Europe.

Cerebras Systems has announced that it will host DeepSeek's R1 artificial intelligence model on its proprietary wafer-scale hardware, achieving processing speeds up to 57 times faster than traditional GPU-based solutions. This deployment delivers 1,600 tokens per second and addresses concerns over data sovereignty by keeping sensitive data within U.S. borders.

NVIDIA has approved Samsung Electronics' fifth-generation high-bandwidth memory (HBM3E) chips for use in its artificial intelligence applications. This approval marks a significant milestone for Samsung, which has been striving to catch up with competitors like SK Hynix in the advanced memory chip market.

Anthropic, the developer of the Claude AI model, is actively working on integrating web search capabilities into its platform. This feature, currently in development and not yet functional, aims to enable Claude to retrieve real-time information from the internet, displaying previews of relevant articles or websites in response to user queries. Such functionality would bridge the gap between Claude's existing static language model and dynamic, real-time data retrieval, aligning it with competitors like OpenAI's ChatGPT, which already offers browsing tools. CEO Dario Amodei has indicated that web integration is a priority for Anthropic, though no specific timeline has been provided for its release.

Microsoft has integrated DeepSeek's advanced reasoning model, R1, into its Azure AI Foundry platform, aiming to enhance AI service offerings for enterprises. This move comes amid concerns about potential intellectual property violations involving OpenAI, a close collaborator of Microsoft. Despite these issues, Microsoft is proceeding with R1's deployment, emphasizing that the model has undergone rigorous safety assessments to ensure reliability. In the coming months, "distilled" versions of R1 will be available for deployment on Copilot+ PCs, enabling developers to leverage its capabilities across various applications.

PwC and Microsoft have announced a strategic collaboration to deploy AI agents across various industries, aiming to enhance efficiency, foster innovation, and drive business growth. PwC's Agents Factory will play a pivotal role in this initiative by ensuring data privacy, ethical AI use, and the swift creation of customized agents tailored to specific client needs and industry requirements.

💡 Emerging AI Technologies and Companies

AI Robotics boom continues with SoftBank’s $500M deal

SoftBank is set to invest $500 million in SkildAI, a rapidly growing AI robotics startup, valuing the company at $4 billion. Founded in 2023, SkildAI specializes in developing adaptable AI models for various robotic applications. This investment reflects the increasing interest in AI-powered robotics, with notable backers like Jeff Bezos also ramping up funding in the sector. Other startups, such as Physical Intelligence and Figure AI, have recently secured substantial investments to advance robotic "brains" and humanoid robots. As demand for intelligent and versatile robots grows across industries, companies like SkildAI are positioning themselves at the forefront of this technological revolution.

Chinese start-up Unitree sees humanoid robots in wide commercial use this decade

Chinese start-up Unitree, renowned for its quadruped robot dogs, anticipates significant advancements in humanoid robot capabilities by 2026, leading to clear commercial applications within three to five years. The company showcased this progress during the recent Spring Festival Gala, where 16 of its H1 humanoid robots performed a fully AI-driven dance alongside human dancers, marking a milestone in large-scale automated performances.

🌍 Global AI Race

Eric Schmidt says DeepSeek marks a ‘turning point’ for the global AI race

Former Google CEO Eric Schmidt has described the emergence of DeepSeek, a Chinese AI startup, as a "turning point" in the global AI race. DeepSeek's R1 model rivals leading AI systems from companies like OpenAI, achieving comparable performance with significantly fewer resources. Schmidt emphasizes that this development demonstrates China's capability to compete with major tech firms using more efficient methods, underscoring the need for increased investment in AI research and development to maintain a competitive edge.

If you want more updates related to AI, subscribe to our Newsletter