AI Daily · Nov 12, 2026: Claude Sonnet 4.5 retires November 30, migrate to 5.5
Claude Sonnet 4.5 has 18 days left before retirement, so migrate API and agent workloads to Sonnet 5.5; Google launched the Gemini Decision 1 scoring API at $0.10/M input with free output, live on OpenRouter and LiteLLM.
Pick a date
-
Claude Sonnet 4.5 to be retired on November 30, migrate to 5.5
A platform notice says Claude Sonnet 4.5 will stop being served on November 30, 2026, with 18 days left, and urges users to migrate API and agent workloads to Claude Sonnet 5.5 and back up data.
-
Google launches Gemini Decision 1 scoring API in AI Studio
Gemini Decisions API, powered by Gemini Decision 1, scores closed-set questions in one pass: yes/no, multiple choice up to 255 options, and probability scores, over text, images, audio, video and PDFs, up to 1,000 questions. It is also live on OpenRouter at $0.10/M input with free output and 262K context.
-
LiteLLM Gateway adds support for Gemini Decision 1 scoring model
LiteLLM Gateway now supports Gemini Decision 1: requests go to /v1/decisions with gemini/gemini-decision-1, questions can be yes/no, multiple choice or scored, and the model returns a probability per answer at $0.10 per 1M input tokens.
-
Perplexity red-teams SPACE sandbox with 12 models
Perplexity published findings from red-teaming SPACE, its VM sandbox for computer agents: 12 models were asked to escape the VM and steal a host credential, and to bypass network policy to reach a blocked URL. It says CTF-style prompts can misdirect agents, security-specific training did not guarantee better results, and after fixing the network flaws no bypasses were found.
-
OpenRouter reports GLM 5.3 Flash latency down 40% after routing update
OpenRouter says its new routing algorithm cut time-to-first-token for GLM 5.3 Flash by 40% over the past month, raised throughput 10%, and lowered cost per 1M tokens by 17%.
-
MiniMax H3 Image open-weight model generates native 4K images
MiniMax H3 Image, a new open-weight image model, is available on each::labs; it produces native 4K images in about 10 seconds, up to 16MP in any aspect ratio, and accepts up to 16 reference images.
-
MiniMax H3 Image and Highspeed models launch on Picsart with 4K output
MiniMax H3 Image and H3 Image Highspeed are live on Picsart, offering 4K generation and editing in about 10 seconds, native ~16MP output at any aspect ratio, and multi-reference editing.
-
NVIDIA Alpamayo 2 Edge 2B driving VLA model available on Hugging Face
NVIDIA released Alpamayo 2 Edge, a 2B-parameter reasoning vision-language-action model for real-time in-vehicle deployment on DRIVE AGX Thor, generating driving trajectories and causal explanations; it can be tried on Hugging Face.
-
LiteLLM rewrites core routing logic in Rust with differential testing
A technical deep dive covers LiteLLM's rewrite of its core routing logic in Rust, explaining the pre-rewrite refactor and how differential testing plus agents are used to catch regressions.
-
Gemini-powered intelligent eyewear opens for pre-order with Samsung and eyewear brands
Google Gemini announced intelligent eyewear built with Samsung, Gentle Monster and Warby Parker; the glasses respond to "Hey Gemini" for real-time translation, calendar checks and local recommendations, and are now available for pre-order.