TLDR AI 2025-11-13

This week, Baseten released Kimi K2 Thinking on Model APIs. Kimi K2 Thinking rivals the leading closed-source agentic models and is engineered for complex reasoning and agentic workflows.With Baseten's API you can get an LLM that is smarter, faster and cheaper. The Kimi K2 Thinking API is built for production workloads with high uptime, scalability, and performance stability:

🧨 Blistering TTFT at 0.3 seconds

⚡140+ tokens per second

🤝 Fully OpenAI-compatible

🛠️ Structured outputs to ensure tool calling quality

Test Kimi K2 Thinking today

OpenAI has released its latest models, GPT-5.1 Instant and GPT-5.1 Thinking, which improve performance and enable more customized behavior.

Anthropic is investing $50 billion to build AI data centers in Texas and New York with Fluidstack. This effort supports the US AI Action Plan and reflects a long-term push for domestic AI leadership.

Marble turns text prompts, photos, videos, or 3D layouts into persistent, editable, and downloadable 3D environments, available now with a free tier. This is the company's first release after raising $230 million and positions the company as the leading “world model” provider for gaming, VFX, VR, and robotics customers.

OpenAI spent $5.02 billion on inference alone with Microsoft Azure in the first half of 2025. This pattern continued through the end of September. The company's inference costs have risen consistently over the last 18 months. OpenAI's inference costs easily eclipse its revenues.

Surprisingly, the authors of "AI 2027" (who predict superintelligence within a decade) and "AI as Normal Technology" (who expect gradual diffusion like electricity) agree on many areas. They say most benchmarks will be saturated by 2027-2028 with AI matching expert humans, yet systems will still regularly fail at mundane tasks like "book me a flight to Paris" through 2029 due to long-tail errors. The world will look basically the same in 2029, even with AGI right around the corner.

The 2025 automation landscape is confusing. Many solutions can deliver quick wins, but also trap value in silos, create technical debt, and block visibility. This guide provides a comprehensive comparison of leading automation technologies and platforms, highlighting what truly matters when orchestrating complex, cross-functional processes. Download the Camunda comparison guide.

ERNIE-4.5-VL-28B-A3B-Thinking is a 28-billion-parameter multimodal AI model that only activates 3 billion parameters during operation. It has a sophisticated routing architecture that allows it to match or exceed the performance of much larger competing systems on several categories of tasks. The model's most distinctive feature is 'Thinking with Images', a capability that allows the AI to dynamically zoom in and out of images to examine details. The feature dramatically elevates the model's ability to handle long-tail visual knowledge. The model has been released under the permissive Apache 2.0 license, which allows unrestricted commercial use.

AlphaProof reached silver medal performance in the International Mathematical Olympiad 2024. This post shares details about how the model works. AlphaProof demonstrates how test-time compute can be scaled to obtain improved answers over many orders of magnitude. The researchers developing it were entirely limited by how many TPUs they could get their hands on.

Software interfaces force their way of thinking on users. AI allows you to bypass all these tools - you just have to feed it the right data, and it can complete complex tasks. Doing this right can result in you getting rid of almost all of your productivity tools. This article shows readers how to get started, along with examples of how AI can completely change a workflow.

Microsoft aims to double its total data-center footprint over the next two years. Its newest AI super factory sits within a set of two-story structures in Atlanta. The site will contain hundreds of thousands of Nvidia GPUs and dedicated high-speed connections to other locations. Part of Microsoft's Fairwater network of AI centers, it spans more than 1 million square feet across 85 acres.

Spinning up AI agents has never been easier, but 62% of teams lose time or revenue to reliability issues. What's the missing piece? Read the 2025 Production AI Stack Report by Temporal

Anthropic divided eight of its researchers into two teams, one with Claude access, one without, and tasked them with programming quadruped robots to fetch beach balls.

Coding agents are very useful, but building a business around them is quite hard.

The platform connects brands with voice rights owners, ensuring permission and fair compensation, and includes the voices of Michael Caine, Mark Twain, and Maya Angelou.

OpenAI is resisting legal efforts by the New York Times to obtain millions of private ChatGPT conversations, citing user privacy and data security risks.