AI Daily · Nov 22, 2026: OpenRouter off-boards three inference providers over data handling
OpenRouter off-boarded three inference providers and pulled many endpoints from ZDR eligibility after data-handling findings, while OpenCode cut GLM 5.3 and DeepSeek V4.1 Flash prices by 60%.
Pick a date
-
OpenRouter off-boards three inference providers after data-handling findings
Security researchers found data-handling problems at some inference providers on OpenRouter; OpenRouter says it was not breached, has off-boarded 3 providers, removed many endpoints from ZDR eligibility, and notified affected customers.
-
OpenCode cuts GLM 5.3 and DeepSeek V4.1 Flash prices by 60%
OpenCode is offering GLM 5.3 and DeepSeek V4.1 Flash at 60% off, claiming prices 60% below Fireworks and Baseten while running on its own GPUs.
-
Thinky Machines releases Inkling 2 Mini, 26B looped transformer under Apache 2.0
Inkling 2 Mini is a 26B-parameter looped transformer small enough to run on a laptop, scoring 20.1 on the AA Index and released under Apache 2.0; the lab also shares a variable-loop research model.
-
Inkling 2 Mini runs locally at 210 tok/sec on M5 Ultra, 1.8x faster than Claude Sonnet 5.5
Hugging Face amplified a report that Inkling 2 Mini, built with ThinkyMachines, decodes at 210 tok/sec and prefills at 3620 tok/sec on a 96GB M5 Ultra Mac Studio, and at 125 tok/sec decode on a 48GB M5. The post claims it is over 1.8x faster than Claude Sonnet 5.5 and 11x cheaper at API pricing.
-
Gemini 4 Neon rolls out to Google AI Pro and Ultra subscribers
Gemini 4 Neon is now available to Google AI Pro and Ultra subscribers on mobile, desktop and web, positioned for frontier-level reasoning with fast responses.
-
Inference providers adopt reserved-tokens-per-hour pricing with overage penalties
A developer notes many inference companies now charge a fixed dollar amount per hour for reserved tokens: usage above the reservation is billed at a much higher rate, usage below is still paid for, and the provider resells the capacity.
-
US House report backs open-source AI over bans on foreign models
A new US House report argues American researchers, startups, nonprofits and hobbyists benefit from open-source AI regardless of origin, and that the answer to Chinese open-source AI is more American open-source AI rather than a ban.
-
Hex dictation app launches on iOS App Store, free and fully local
Anomaly's Hex dictation app is now on the iOS App Store, running fully locally and offered free forever.
-
LiteLLM Lens CLI compares models and agent harnesses from terminal
LiteLLM Lens CLI lets developers compare model and agent-harness combinations (Claude Agent SDK, OpenAI Agents SDK, LangGraph, custom agents) from the terminal instead of hand-building eval loops and spreadsheets.