Good Morning, Model Builders!
DeepSeek is pushing down coding costs, Anthropic is spending big on compute, and OpenAI may have new image models on the way.
OpenAI's Aurora Image Models Surface

TL;DR: Three unannounced OpenAI models, aurora-red-alpha, aurora-green-alpha and aurora-blue-alpha, briefly appeared in the company's API before showing up in Arena image-generation tests. Early testers report differences in speed and image quality, fueling speculation that OpenAI is preparing GPT-Image-3. The company has not confirmed the models or announced a release date. Read more →
DeepSeek Flash Tops Open Code Arena

TL;DR: Arena reports DeepSeek-V4.1-Flash-20261017 (Max) is now the top open model in Code Arena WebDev, ahead of Kimi K3 (Max), and 12th overall with 1670 points. It is priced at $0.3 input and $1.2 output per million tokens, and gained 52 points over the September Flash build. Read more →
Anthropic Seeks $60B for Google TPUs

TL;DR: Anthropic is seeking roughly $60 billion in financing for Google TPU capacity, including a proposed $9 billion investment in junior debt from Blackstone-managed funds. The debt carries a 10% coupon, with total yields exceeding that level after fees and discounts. The deal highlights the soaring cost of AI infrastructure and Anthropic's growing reliance on Google silicon. Read more →
Nvidia Pushes TSMC 3nm for Custom HBM

TL;DR: Nvidia has reportedly asked Samsung, Micron and SK Hynix to manufacture base dies for its next-generation custom HBM using TSMC's 3nm process, according to South Korean media. The move could make memory supplies more interchangeable across manufacturers as HBM prices rise, while giving TSMC a greater role in the AI memory supply chain. Read more →
Claude Opus 5.5 Silently Falls Back

TL;DR: Users report that Claude Opus 5.5 sessions keep quietly switching to Opus 4.8. XDA Developers reports the fallback is a deliberate mechanism meant to prevent a security leak. For anyone paying for Opus 5.5, it means the model answering a request may not be the one they selected. Read more →
Replit Swaps Default Model to OpenAI

TL;DR: Replit CEO Amjad Masad clarified over the weekend that OpenAI will replace Anthropic as the coding platform's default model provider. Claude remains available to customers who choose it. The switch moves one of the largest AI coding platforms' default traffic from Claude to OpenAI models. Read more →
Google Open-Sources Harness2 Agent Framework

TL;DR: Google Research released Harness2, an Apache-2.0 framework that rewrites an agent's harness at test time instead of retraining the model. One refinement step lifted GLM-5 by 11.5 points on WorkBuddy-Bench with no weight updates, a cheap way to get more out of the models teams already run. Read more →

Top AI models. Lower costs.
Access GPT, Claude, Gemini, and more through one OpenAI-compatible API—and save up to 70% on AI API costs. Create your DIT account, get your API key, and start building.