AI
-
New Products
Cohere opens Model Vault to third-party open models, from DeepSeek V4.1 Flash to Gemma 4 and GPT-OSS
Cohere now lets enterprises run leading open models next to its own Command, Embed and Rerank families inside Model Vault, its single-tenant managed inference service with zero data retention by default. A smaller set of…
-
New Products
Apple puts Audio Intelligence for Apple Watch into developer beta, with Live Rewind and Siri Recap listening to conversations
Apple's watchOS 27.3 developer beta brings the Audio Intelligence features it announced in September to developers, including Live Rewind, which shows the last 15 seconds of speech, and Siri Recap, which writes short sum…
-
Agents
Apple opens Siri AI to every developer in the 27.3 betas, letting apps expose tools through App Intents and teach multi-step tasks with app skills
A Siri AI Developer Preview in the iOS, iPadOS, macOS and visionOS 27.3 betas lets any app make its own actions and content available to Siri AI, which until now worked mainly with Apple-defined schemas. Developers can a…
-
New Products
Google rolls out Guided Learning in Classroom worldwide and says Gemini and Search will be off for K-12 student accounts until schools opt in
Google made its answer-withholding Guided Learning tutor available globally in Google Classroom on November 3 and will pilot a version in Search for select U.S. schools. It also said Gemini, Gemini Notebook, Gemini in Cl…
-
New Products
Anthropic launches message threads in public beta, letting Claude agents send only new turns instead of the whole conversation
Anthropic's new message threads beta keeps a conversation on its servers for a few minutes so that each later Messages API call carries only what changed. The company says this cut median time to first token by about 35%…
-
New Models
Google deprecates Gemini 2.5 Pro and 2.5 Flash in the Gemini API and will automatically reroute their traffic to Gemini 3.8 Flash and 3.5 Flash-Lite on December 1
Google has pulled Gemini 2.5 Pro and Gemini 2.5 Flash from the Google AI Studio model picker and says that from December 1, 2026 any API call naming either model will be served, and billed, by a newer model. Developers w…
-
Research
Anthropic releases FlashProve, an open-source system that cryptographically proves which model produced a response, running a 400B-parameter model on one H100
Anthropic and academic collaborators have published FlashProve, a research prototype that lets a model provider commit to hidden weights and then attach a zero-knowledge proof showing each reply came from that exact mode…
-
Safety & Ethics
OpenAI raises its security bug bounty rewards, paying up to $30,000 for critical flaws in ChatGPT, Codex and the API and about $172,000 for top findings with bonuses
OpenAI says it is more than quadrupling, on average, the midpoint of its reward ranges for medium, high and critical vulnerabilities, and adding bonus multipliers for report quality, researcher loyalty and priority campa…
-
Research
MIT CSAIL researchers report GPT-6 Astra can read and directly rewrite neural network weights, lifting a small Qwen model's MATH-500 score from 7.9% to 48.1%
In a November 2 research post, four MIT CSAIL researchers describe what they call Weight-Space Programming: a frontier LLM inspecting and editing a network's raw weights without gradient training. Their best result fixed…
-
Safety & Ethics
Anthropic's amended IPO filing warns it may soon meet its own threshold for AI-accelerated research, as third-quarter revenue reaches $17 billion
In Amendment No. 1 to its S-1, filed on November 2, Anthropic says internal estimates for future models indicate it may imminently meet its Responsible Scaling Policy definition of dramatically accelerated progress attri…