Here’s what moved AI forward this week

OpenAI has reportedly entered an internal “Code Red” phase as competition in advanced AI reasoning intensifies. According to internal signals, a new model codenamed “Garlic” is performing strongly in coding and reasoning tasks, rivaling top-tier models like Gemini 3 DeepThink and Claude Opus.

Leadership has hinted at an accelerated release timeline, suggesting we could see a major OpenAI model update sooner than expected. If accurate, this could mark the next leap in frontier reasoning models.

The pace of model releases is accelerating, and OpenAI appears unwilling to concede ground in advanced reasoning performance.

Source

Chinese AI company Cling AI released a rapid series of video models this week, including:

While still inconsistent in precision, these models show rapid progress in video manipulation, audio sync, and avatar realism.

Global competition in generative video is intensifying, with China closing the gap faster than expected.

Source

Runway unveiled early demos of Gen 4.5, a next-generation video model that appears to outperform current public models in motion consistency, realism, and creative complexity.

Although not yet available, early benchmarks and demos suggest it could rival or surpass Sora and Veo 3.1 once released.

Video generation quality is improving faster than anticipated, reshaping filmmaking, marketing, and content creation.

Source

Google officially released Gemini 3 DeepThink to Ultra subscribers. This model tops multiple reasoning benchmarks, including Humanity’s Last Exam and ARC-AGI 2.

Despite its strength, early users report limited availability due to high demand, even at the $250/month tier.

Deep reasoning models are becoming a premium tier, signaling a split between everyday AI and elite cognitive systems.

Source

At AWS re:Invent, Amazon announced:

Amazon is positioning itself as the backbone for enterprise AI development.

The AI race is no longer just about models it’s about infrastructure dominance.

Source