Microsoft takes on Claude with seven models costing up to 95% less for coding tasks

Microsoft takes on Claude with seven models costing up to 95% less for coding tasks

Key TakeawaysMicrosoft launched seven MAI models covering coding, cybersecurity, image editing, speech, transcription and decision-making.MAI Cyber 2 Flash claims GPT-6 Astra-level quality at less than 2% of the cost per task, according to Microsoft.Microsoft's Maia 200 accelerator is in production, running its largest models and delivering 1.7 times more tokens per watt, the company said. Microsoft has unveiled seven new models under its MAI family, targeting coding, cybersecurity, image editing, speech, transcription, and automated decision-making as it looks to compete with OpenAI and Anthropic on performance and cost. Announced on November 17 at its Ignite conference, the lineup includes MAI Cyber 2 Flash, MAI Code 2, MAI Code 2 Flash, MAI Image 2.7, MAI Realtime 1, Dragon Copilot Transcribe and Decision 1.1. The launch reflects Microsoft’s push to make advanced models more affordable for businesses running automated agents continuously. The company said rising token costs are making price-performance increasingly important for enterprises looking to scale these systems. Microsoft is positioning the models for specific workloads rather than relying on a single system for every task. It also highlighted Maia 200, its second-generation AI accelerator, which is running its largest models in production. Microsoft targets lower costs MAI Cyber 2 Flash is designed for continuous cybersecurity monitoring and defensive work, including identifying and patching vulnerabilities in live production systems. Microsoft claims it matches GPT-6 Astra’s quality at less than 2 percent of the cost per task on the CyberGymE2E benchmark. The company said the model builds on earlier MAI Cyber systems used by defenders. Its predecessor accounts for almost 40 percent of scans in MDASH, Microsoft’s security assessment platform, while 90 percent of scans that began with MAI continued using it. For software developers, MAI Code 2 targets everyday enterprise coding tasks. Microsoft said it delivers 98 percent of GPT-6.1 Sol’s coding performance at about 40 percent of the price. Against Anthropic’s Claude Opus 5.5, it claims 96 percent of the performance at just 5 percent of the cost. A lighter version, MAI Code 2 Flash, is aimed at routine tasks such as code explanations, targeted edits and fixes. It is available in GitHub Copilot and in limited preview through Microsoft Foundry. Models span image and speech Microsoft also introduced MAI Image 2.7, which it says ranks first in image editing on Artificial Analysis. The company claims it outperforms OpenAI’s GPT Image 2.6 Sunburst at one-fifth of the price. Its text-to-image generation performance ranks fourth on the same evaluation. MAI Realtime 1 is Microsoft’s first speech-to-speech model, built for natural conversations with low latency. It starts at $0.01 per audio minute. Dragon Copilot Transcribe, meanwhile, is based on MAI Transcribe 2 and powers 10 million patient encounters each month, according to Microsoft. The company said it has halved multilingual transcription error rates. Decision 1.1 provides model routing within applications, directing tasks to larger models only when needed. Microsoft said it is 50 times faster and 79 times cheaper than GPT-Sol 6.1 and is used by 4,000 customers. The company is also introducing MAI Licensed Weights, allowing businesses to customize models with proprietary data and run them in their own Azure environments. Alongside the models, Microsoft highlighted Maia 200, its second-generation accelerator for AI workloads. The chip is in production, running Microsoft’s largest models and delivering up to 1.7 times more tokens per watt than leading GPU infrastructure, according to the company. By combining specialized models, customizable weights, and in-house accelerators, Microsoft aims to give businesses more control over how they deploy and pay for automated systems. The announcement comes as major technology companies compete to lower the cost of running increasingly capable models.Get the latest in engineering, tech, space & science - delivered daily to your inbox.With over a decade-long career in journalism, Neetika Walter has worked with The Economic Times, ANI, and Hindustan Times, covering politics, business, technology, and the clean energy sector. Passionate about contemporary culture, books, poetry, and storytelling, she brings depth and insight to her writing. When she isn’t chasing stories, she’s likely lost in a book or enjoying the company of her dogs.

Original Source

Read the full article at Interestingengineering →

KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.