Exciting News! Microsoft Azure AI Launches GPT-4o Mini with Lower Costs and Faster Speeds

AIbase基地

Published inAI News · 4 min read · Jul 19, 2024

222

OpenAI announced today that its new flagship model, GPT-4o mini, has been launched on Azure AI, supporting text processing capabilities, with plans to introduce image, audio, and video functionalities in the future.

GPT-4o mini is significantly smarter than GPT-3.5Turbo—scoring 82% on the Massive Multitask Language Understanding (MMLU) benchmark compared to GPT-3.5Turbo's 70%—and is over 60% cheaper. This model offers an expanded 128K context window and integrates the enhanced multilingual capabilities of GPT-4o. The Azure OpenAI Studio Playground allows free trials of GPT-4o mini.

Microsoft Azure AI brings default security to GPT-4o mini, along with extended data residency and service availability upgrades. Customers can expect enhanced performance and features on Azure AI, particularly for streaming scenarios such as assistants, code interpreters, and retrieval.

Azure AI has announced global pay-as-you-go options with the highest throughput limits for GPT-4o mini. Now, customers can pay flexibly based on consumed resources, with traffic routed globally for higher throughput, and have control over the storage location of their data. The global pay-as-you-go deployment option offers a throughput of 15 million tokens per minute (TPM) and 99.99% availability for GPT-4o mini, at the same industry rates as OpenAI.

GPT-4o mini will be available on Azure AI this month and will be offered in Batch services. Batch delivers high-throughput jobs at a 50% discount rate within 24 hours by utilizing off-peak capacity. This is only possible with Microsoft running on Azure AI, enabling Microsoft Azure AI to provide off-peak capacity to customers.

Microsoft Azure AI will also release fine-tuning capabilities for GPT-4o mini this month, allowing customers to further customize the model according to specific use cases and scenarios. Following last month's update to token-based training billing, Microsoft Azure AI has reduced hosting costs by 43%. Combined with its low inference prices, this makes Azure OpenAI service fine-tuning deployments the most cost-effective option for customers with production workloads.

Key Highlights:

⭐ GPT-4o mini launches on Azure AI, supporting fast and comprehensive text processing functionalities

⭐ The new model is smarter than its predecessor, over 60% cheaper, and offers a broader context window and multilingual capabilities

⭐ Azure AI provides a global pay-as-you-go deployment option for GPT-4o mini, offering high throughput and 99.99% availability

Meta's Latest Audio Model SPIRIT LM: Making AI Not Just Talk, But Also Express Emotion!

Recently, Meta AI open-sourced a foundational multimodal language model named SPIRIT LM, which can freely mix text and speech, opening new possibilities for multimodal tasks involving audio and text. SPIRIT LM is based on a pre-trained text language model with 7 billion parameters, which has been continuously trained on text and speech units, expanding into the speech modality. It can understand and generate text like a large text model, while also being capable of understanding and generating speech, and even mixing text and speech to create various forms of expression.

Samsung Launches New Gauss 2 AI Model, Expected to Be the Next Galaxy Brain

At this year's Samsung Developers Conference, Samsung officially unveiled the new Gauss 2 AI model, marking another significant advancement in its artificial intelligence endeavors. The Gauss 2 model upgrades in performance and efficiency, with applications ranging from smartphones to home appliances, encompassing a variety of smart devices. Gauss 2 has multimodal capabilities, allowing it to process images, text, and code simultaneously, enhancing the integration of AI into various device ecosystems. Samsung has developed three versions: Compact, Balanced, and

Google's New Gemini Beta Reclaims Top Spot in AI Rankings, GPT-4o Only Reigns for a Day

Recently, the competition between Google and OpenAI has intensified again. Just one day after the new GPT-4o topped the AI rankings, Google launched its latest beta model, Gemini-Exp-1121, quickly regaining the championship title. Just a week prior, Google had released Gemini-Exp-1114, indicating a rapid response from Google to OpenAI's developments. Jack Rae, the Chief Scientist at Google DeepMind, stated that

Next-Generation Visual Tracking Model SAMURAI: Tracking Subjects in Complex Scenes

Recently, a research team from the University of Washington released a new visual tracking model called SAMURAI. This model is based on the Segment Anything Model2 (SAM2) and aims to address the challenges encountered during visual object tracking in complex scenes, particularly when dealing with fast-moving and self-occluding objects. While SAM2 excels in object segmentation tasks, it has some limitations in visual tracking. For instance, its fixed-window memory approach fails to consider objects in crowded scenes.