API
< Back to all clusters
[TECHNOLOGY] · United States · 7 sources

started · updated

Microsoft rolls out new AI models for imaging, coding and cybersecurity

Microsoft announced the replacement of OpenAI image models with its own MAI‑Image‑2.5 in core products such as OneDrive and PowerPoint, estimating an 85 % reduction in image‑generation costs.

The company also released MAI‑Code‑1‑Flash, a smaller coding assistant that recorded a 10 % higher acceptance rate than Claude Haiku 4.5 and 7 % higher than GPT‑5.4 Mini, while using 11–13 % fewer tokens per task.

In the security arena, Microsoft unveiled MAI‑Cyber‑1‑Flash, a specialist AI model that can handle about 90 % of vulnerability‑detection work at roughly half the cost of its larger counterpart. In internal testing the combined MDASH system (MAI‑Cyber‑1‑Flash plus GPT‑5.4) achieved a 96 % score on the CyberGym benchmark, surpassing competing models from Anthropic, Google and OpenAI. The model will power the Project Perception platform, slated for public preview on 3 August, which automates vulnerability identification and remediation across Microsoft and third‑party products.

Entities

Hayete Gallot · MAI-Cyber-1-Flash · MAI‑Code‑1‑Flash · MAI‑Cyber‑1‑Flash · MAI‑Image‑2.5 · Microsoft · Microsoft Corporation · Project Perception · Satya Nadella

Claims

What the coverage asserts, and how many sources carry each claim.

  • [● 3 SOURCES] The MDASH system combining MAI‑Cyber‑1‑Flash with GPT‑5.4 scored 96 % on the CyberGym benchmark, beating Anthropic’s Mythos 5, Google’s 3.5 Flash Cyber and OpenAI’s GPT‑5.5 Cyber. 4sysops.com · asiaone.co.in · memeburn.com
  • [● 3 SOURCES] MAI‑Cyber‑1‑Flash handles about 90 % of security tasks at roughly half the cost of larger models. 4sysops.com · asiaone.co.in · memeburn.com
  • [○ 1 SOURCE] Project Perception, scheduled for public preview on 3 August, integrates the new AI models to automate vulnerability detection and remediation. asiaone.co.in
  • [● 3 SOURCES] Microsoft’s rollout of proprietary AI models reflects a strategy to reduce reliance on external providers like OpenAI. 4sysops.com
  • [○ 1 SOURCE] MAI‑Code‑1‑Flash used 11 % fewer tokens than Claude Haiku 4.5 and 13 % fewer than GPT‑5.4 Mini. 4sysops.com
  • [○ 1 SOURCE] Microsoft replaced OpenAI image models with MAI‑Image‑2.5 in OneDrive and PowerPoint, reducing image‑generation costs by 85 %.
  • [○ 1 SOURCE] MAI‑Code‑1‑Flash achieved a 10 % higher acceptance rate than Claude Haiku 4.5 and 7 % higher than GPT‑5.4 Mini. 4sysops.com