Microsoft Says Newest AI Fashions Beat Claude, Google's Nano Banana – Decrypt

On the primary day of the annual Microsoft Construct occasion on Tuesday, the Home windows developer unveiled seven new AI fashions, claiming they outperformed Anthropic’s Claude Sonnet 4.6 and Google’s Nano Banana 2 in blind testing and image-editing benchmarks.

The declare comes as Microsoft makes an attempt to determine itself as a frontier AI developer relatively than solely OpenAI’s largest backer and infrastructure supplier.

“Tremendous excited to announce seven new world-class MAI fashions immediately,” Microsoft AI CEO Mustafa Suleyman wrote on X. “They symbolize what we take into account a brand new period in AI designed to maintain you in management and on the frontier.”

On the heart of the discharge is MAI-Considering-1, a reasoning mannequin that Microsoft describes as its flagship textual content basis mannequin.

In response to Suleyman, MAI-Considering-1 was most well-liked over Anthropic’s Claude Sonnet 4.6 in blind exams carried out by unbiased evaluators. He added that the mannequin scored 97% on AIME 2025, a benchmark that measures superior problem-solving and reasoning abilities.

Suleyman stated the SWE Bench Professional end result locations the mannequin “proper alongside Opus 4.6 on one of many hardest coding benchmarks.”

The corporate additionally launched MAI-Code-1-Flash, a light-weight coding mannequin constructed for GitHub Copilot and Visible Studio Code; MAI-Picture-2.5 and its Flash variant, which Microsoft says outperform Google’s Nano Banana Professional on image-editing duties; MAI Transcribe-1.5, a transcription mannequin that helps 43 languages; and MAI-Voice-2, a speech-generation mannequin able to producing natural-sounding voices in 15 languages and adapting to a speaker from a brief audio pattern.

“That is a rare time in expertise. The compute used to coach frontier fashions has elevated by an element of 1 trillion,” Suleyman stated in a separate weblog submit asserting the brand new fashions. “Now we count on one other thousand-fold enhance over the following three years, which in flip means extra superior capabilities, and the continued rollout of ever simpler AI.”

The announcement comes as competitors amongst main AI builders continues to accentuate.

Final week, Anthropic introduced the launch of its newest flagship mannequin, Opus 4.8, which the corporate stated is quicker and smarter on benchmark exams and comes with a set of recent options. On Tuesday, Anthropic introduced an growth of its Challenge Glasswing, giving 150 corporations entry to its new cybersecurity-focused Mythos mannequin.

In the meantime, at Google I/O in Might, Google unveiled Gemini Omni, a multimodal AI mannequin that mixes Gemini with the corporate’s Veo, Nano Banana, and Genie media-generation fashions, alongside Gemini Spark, a cloud-based AI agent designed to handle duties throughout apps and workflows on a person’s behalf.

Microsoft’s new mannequin launch suggests a broader effort to construct proprietary AI programs because it expands past its longstanding reliance on OpenAI expertise, saying that MAI “delivered the best win charge, outperforming GPT-5.5 on high quality, whereas being 10x decrease on price.”

“Builders and companies have been crying out for AI that delivers on their phrases and beneath their say,” Suleyman wrote. “We see this as a significant step in direction of delivering that.”