vLLM trades up to 18% slower output for a bigger KV cache
vLLM now sizes its KV cache from free GPU memory by default, removes Model Runner V1 and raises the default block size from 16 to 64.
vLLM now sizes its KV cache from free GPU memory by default, removes Model Runner V1 and raises the default block size from 16 to 64.
Google said Nov. 16, 2026, that the Gemini side panel in Apps Script is generally available, with a gradual rollout and admin controls.
OpenAI's GPT-6.2 Astra is listed at $8 per 1M input tokens, down from $10, and the API now drops reasoning items when a conversation prefix changes.
Alibaba made Qwen4-Max generally available through its API at $1.50 per 1M input tokens and $4.50 per 1M output, with thinking on by default.
Tencent released Hy4 flash, its cheapest Hy model, at about $0.070 per 1M input tokens and $0.281 per 1M output, free on WorkBuddy through Nov. 30.
OpenAI's ChatGPT Wallet lets its dots agents complete purchases for a limited number of U.S. Pro subscribers, with a biometric approval on each one.
Google made Gemini 4 Hydrogen generally available at $1.50 per 1M input tokens and $7.50 per 1M output, double the current Gemini 3.8 Flash rate.
DeepSeek made web search, URL reading, file input and safety checks from its app available through the API, with per-use tool fees on top of tokens.
llama.cpp and ggml reached 1.0.0 and now mark every breaking change to their interfaces with a major-version bump, after deleting the deprecated public API.
Google said Nov. 10, 2026, that skills in Workspace are generally available and rolling out gradually, with skills on by default and activity logged for admins.
Thinking Machines made the Inkling 2 Small weights downloadable under Apache 2.0, and the main checkpoint needs at least 500GB of GPU memory.
Anthropic made Claude Fable 5.2 generally available at $8 and $40 per million tokens, 20% below Fable 5.1, with stricter rules that return errors.