OpenAI has initiated a limited preview of its next-generation GPT-5.6 series, comprising Sol, Terra, and Luna models. These models demonstrate significantly enhanced agentic capabilities, particularly across coding, biology, and cybersecurity domains. GPT-5.6 Sol, positioned as the flagship model, achieves state-of-the-art performance on benchmarks like Terminal-Bench 2.1, GeneBench v1, and ExploitBench.

New features include a "max reasoning effort" for Sol to enable deeper reasoning and an "ultra mode" that leverages subagents to accelerate complex tasks. Concurrently, OpenAI has implemented its most robust, layered safeguard stack to date, designed to mitigate misuse while preserving legitimate access for defensive work.

These safeguards encompass model-level refusals, real-time misuse classifiers, account-level reviews, and differentiated access. OpenAI dedicated over 700,000 A100-equivalent GPU hours to automated red-teaming to identify and address universal jailbreak attacks. The GPT-5.6 series is slated for broader availability in the coming weeks, with Sol priced at $5 per 1M input tokens and $30 per 1M output tokens.

👉 Read the full article on OpenAI

Anthropic has accused operators linked to Chinese giant Alibaba and its Qwen lab of orchestrating the largest known "distillation attack" against its Claude AI model. Between April and June 2026, approximately 25,000 fraudulent accounts allegedly generated 28.8 million interactions to extract advanced capabilities from Claude, including agentic reasoning, programming, and long-task resolution.

While "distillation" in AI is a legitimate technique to train smaller, cheaper models using responses from more powerful ones, Anthropic asserts that this instance was conducted on a massive, unauthorized scale, violating terms of use. This incident underscores the escalating technological, economic, and geopolitical competition within the AI sector.

Anthropic previously identified similar campaigns by other Chinese labs (e.g., DeepSeek, Moonshot AI, MiniMax) in February, making the accusation against Alibaba even larger in scale. The U.S. government is concerned that China copying U.S. model capabilities could erode AI leadership, prompting Anthropic to urge Congress to facilitate information sharing without antitrust conflicts, strengthen export controls on advanced chips, and sanction labs engaging in unauthorized distillation.

👉 Read the full article on Gizmodo en Español

This TechCrunch article explores the growing trend of major technology companies, including OpenAI and SpaceX, developing their own custom AI chips to reduce reliance on established suppliers like Nvidia. While the full content of the article was inaccessible, this strategic shift is understood to be driven by the desire to optimize performance for specific AI workloads, lower operational costs, and gain tighter control over their supply chains for AI model training and inference.

Many companies are investing in designing custom silicon to meet the unique demands of their AI applications, which generic GPUs may not fully address. This trend signifies a significant shift in the AI hardware landscape, intensifying competition and potentially challenging Nvidia's dominant position.

👉 Read the full article on TechCrunch