AI News Summary — Week 16.2, 2026

Happy Friday! This week's AI news brings us updates from all the major players: Anthropic retakes the LLM crown with Opus 4.7, OpenAI fires back with a cybersecurity-focused model, and Microsoft continues its march toward AI independence with a new efficient image generation model.

Anthropic Releases Claude Opus 4.7

Anthropic has released Claude Opus 4.7, its most powerful language model yet, narrowly retaking the lead over OpenAI's GPT-5.4 on key benchmarks.

The new model shows significant improvements in agentic coding, long-form reasoning, and autonomous task execution. On the GDPVal-AA knowledge work benchmark, Opus 4.7 achieved an Elo score of 1753, surpassing both GPT-5.4 (1674) and Gemini 3.1 Pro (1314).

Key Improvements

• Agentic Coding (SWE-bench Pro): Resolved 64.3% of tasks, up from 53.4% in Opus 4.6
• Graduate-Level Reasoning (GPQA Diamond): Reached 94.2%
• Visual Reasoning: Improved to 91.0% with tools (up from 84.7%)
• High-Resolution Multimodal: Now processes images up to 2,576 pixels on their longest edge — a three-fold increase

The model introduces a new "effort" parameter allowing users to select xhigh (extra high) effort for more granular control over reasoning depth. Anthropic also launched "task budgets" in public beta, letting developers set hard ceilings on token spend for autonomous agents.

Opus 4.7 is available across Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, with API pricing held steady at $5/$25 per million tokens.

The Rivalry Heats Up

The race between AI giants is tighter than ever. On directly comparable benchmarks, Opus 4.7 only leads GPT-5.4 by 7-4. GPT-5.4 still leads in agentic search (89.3% vs 79.3%), multilingual Q&A, and raw terminal-based coding.

OpenAI Launches GPT-5.4-Cyber

Just a week after Anthropic's Mythos announcement, OpenAI unveiled GPT-5.4-Cyber, a specialized model designed to give cybersecurity teams an edge over AI-augmented attackers.

The release marks a strategic shift toward "cyber defense" — a direct response to Anthropic's Project Glasswing initiative. While details are limited, the model is being positioned as a defensive tool for security professionals.

This comes on the heels of OpenAI's GPT-5.4 release in early March, which brought native computer-use capabilities. On the OSWorld-Verified benchmark, GPT-5.4 achieved a 75% success rate in desktop navigation — up from 47.3% in GPT-5.2 and exceeding average human performance.

Microsoft's MAI-Image-2-Efficient

Microsoft released MAI-Image-2-Efficient, a streamlined version of its flagship image generation model, as the company continues its push for AI independence from OpenAI.

The new model delivers four-times faster throughput on Nvidia H100 processors and is 22% faster than the flagship MAI-Image-2. It also offers 41% cheaper output pricing ($19.50 vs$33 per million tokens), making it suitable for "assembly line" tasks like UI mockups, product photography, and marketing assets.

Microsoft's AI Independence Strategy

The release comes amid growing tension between Microsoft and OpenAI. A leaked internal memo from OpenAI's new CRO suggested the partnership "has limited our ability to meet enterprises where they are," pointing to Amazon Web Services as a key growth driver.

Microsoft has been actively diversifying its AI infrastructure, adding OpenAI to its official list of competitors in mid-2024. The MAI (Microsoft AI) team, led by Mustafa Suleyman, is delivering on the company's strategy to reduce reliance on OpenAI.

The original MAI-Image-2 claimed the No. 3 spot on Arena.ai's leaderboard for its photorealism and complex typography rendering — areas where many image models struggle.

What's Hot This Week

• The LLM race is razor-tight: Opus 4.7 barely edges GPT-5.4 (7-4 on benchmarks) — expect more incremental releases
• Cybersecurity becomes a battleground: Both Anthropic (Mythos) and OpenAI (GPT-5.4-Cyber) are launching specialized security models
• Microsoft goes solo: MAI-Image-2-Efficient is another step toward AI independence from OpenAI
• Agentic AI matures: Task budgets and effort parameters signal AI is becoming a production line item requiring fiscal guardrails

That's a wrap for this week's AI news. Have a great weekend!

Sources: TechCrunch, VentureBeat, SiliconANGLE, Anthropic Blog, OpenAI Blog, Microsoft Tech Community