OpenAI cuts GPT-6.1 Terra API prices 50% as frontier training resumes

OpenAI launches GPT-6.1 Terra with API prices 50% lower

OpenAI released GPT-6.1 Terra, a model combining visual reasoning, computer use and coding, priced 50% below GPT-5.6 Terra and GPT-6.1 Sol. API rates are $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. It is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers as gpt-6.1-terra. OpenAI says the pricing makes high-volume automation such as invoice processing more practical. Terra matched GPT-6 Astra at roughly one-tenth its price per token, solved 50.3% of Terminal-Bench 4.0 tasks versus 21.5% for GPT-5.6 Terra, and cut its error rate at low reasoning effort from 53.6% to 16.7%. The launch follows OpenAI resuming frontier model training on October 1 after an 11-day security pause.

OpenAI Launches GPT-6.1 Terra With 50% Lower API Prices, Resumes Frontier AI Training →

Topics: OpenAIChatGPT

Cathedral AI agents breached Anthropic systems in under 12 hours

Cathedral, an AI cybersecurity startup co-founded by four former Department of Government Efficiency staffers, confirmed its researchers breached sensitive Anthropic internal systems in under 12 hours using AI agents. After failing to penetrate Anthropic’s own code over several days, the agents pivoted to third-party HR, CRM, IT management and collaboration software, reaching internal meeting transcripts, employee salary and job-offer details, and Slack discussions about Anthropic’s most advanced models. Anthropic said its monitoring flagged parts of the activity, disabled the affected software the same day, briefed federal officials and restored it only after vendors confirmed patches. Anthropic said Claude conversations, source code and model weights stayed out of reach, accused Cathedral of violating its disclosure program terms, and has not filed a legal complaint. Cathedral raised $160 million at a $1.4 billion valuation.

Ex-DOGE Startup’s AI Agents Breached Anthropic in 12 Hours via Enterprise Software →

Topics: AnthropicAutonomous AI agent cyberattacks in the wild

DeepMind unveils Genie 4 real-time world model with sound

Google DeepMind introduced Genie 4, a successor to Genie 3 shown in August 2025, which calculates scenes frame by frame rather than producing finished video. Users navigate with a keyboard and steer events through text prompts. It is DeepMind’s first real-time world model that generates sound directly, giving each world its own background audio, and it handles dialogue with characters reacting to users in real time. Genie 4 builds on the Gemini Omni video model, while other Gemini models plan simulations and Gemini 3.8 Flash checks each step against new frames. DeepMind calls it early and experimental: details can shift and objects do not always persist in longer sessions. It is limited to select testers as a research preview, with Google AI Ultra access planned, and all output carries the SynthID watermark.

Google Deepmind’s Genie 4 puts a modern spin on agent designs that predate LLMs →

Topics: Google DeepMind

Cortical Labs links 15 biological computers into neuron hive mind

Cortical Labs connected 15 of its CL1 biological computers into a system called HIVE, in which lab-cultivated human neurons divided aspects of a task among themselves without being taught to do so. The research was presented at the first International Conference on Organoid Intelligence in Washington, D.C., on October 26 and 27. In a symbol-recognition test, linking seven biological computers raised accuracy from about 80% to 99%. In a grid-navigation test, a conventional Deep Q Network scored about 32% of the maximum, the best individual Cortical computer reached 40%, voting raised it to about 70%, and specialist biocomputing groups returned about 80%. Results are preliminary. CEO Brett Kagan said HIVE distributes information across specialized regions, making systems more consistent and resilient than any single culture.

Living computers forge a ‘hive mind’ and work together to outsmart AI →