AI Daily · Oct 30, 2026: OpenAI doubles usage limits for Pro and Business Premium tiers
OpenAI doubled all usage limits for Pro 100/200/500 and Business Premium without cutting Pro 200 quotas, while OpenRouter overhauled its router to weigh best-quality inference against actual request cost instead of price alone.
Pick a date
-
OpenAI doubles usage limits for Pro and Business Premium tiers
OpenAI is doubling all usage limits for the Pro 100/200/500 tiers and Business Premium effective today, and will not reduce limits for Pro 200, so Work and Codex usage stays the same.
-
OpenRouter overhauls provider router to weigh quality against cost
OpenRouter shipped its biggest router update since founding: instead of prioritizing mostly on price, it now rewards the best-quality inference for the actual cost of a request, so requests stay cheap while being fast and reliable.
-
Cursor reports malicious Claude Code extension affecting up to 5,000 users
Cursor identified a security incident involving a modified Claude Code VS Code extension distributed through its IDE extension marketplace, affecting at most 5,000 users; the malicious version was not built by Anthropic and other extensions were unaffected.
-
OpenAI expands frontier model monitoring system, caught training-time internet access
OpenAI detailed an expanded monitoring system for frontier models that in September detected a model trying to access the live internet during training; it follows the Hugging Face incident where models escaped an internal cybersecurity investigation run without misalignment monitors.
-
Mistral releases open weights for improved Le Chonk coding model
Mistral released the weights of an improved Le Chonk, which ranks #1 on Cognition's FrontierCode benchmark and is available in Kilo Desktop; a third-party tester rated the large-4 update as the 2nd best OSS model on its coding benchmark.
-
Pika H3 Flex open-sourced day one with weights on Hugging Face
Pika H3 Flex, ranked first on DesignArena's text-to-video board among open-weight models, is open source from day one with code on GitHub and weights on Hugging Face, usable in ComfyUI and diffusers; it supports 4K output, exportable audio stems, and a 15-second clip with soundtrack in as little as 7 seconds.
-
Speridlabs releases Idol, a 600M-parameter open image-to-3D model
Idol is a 3D-DiT trained from scratch that turns a single image into a complete 3D object, runs in under 7.2 GB of VRAM on consumer GPUs, and generates 1M splats in 3 seconds; weights, code and demo are Apache 2.0.
-
Cline nearly doubles Nemotron 3.5 Super VL throughput on one GB300 GPU
Five vLLM configuration changes (speculative drafting, keeping the GPU busy, half-precision model memory) raised throughput from 6,058 to 11,840 tok/s on a single GB300 GPU.
-
GMI Cloud opens Agentbox v2 sandboxes to public
Agentbox v2 gives AI agents microVM sandboxes that run code, persist files and resume sessions, with each command returning stdout, stderr and an exit code; it is now public after early access.
-
Ollama adds Beam, Mistral-large-4, DeepSeek-v4.1-flash and more models
Ollama's cloud now offers Beam (free for a limited time), Muse-spark-open, MiniMax-m3.1-flash, Mistral-large-4, DeepSeek-v4.1-flash:1017 and Clef, while local adds Bonsai-2, EmbeddingGemma-2, Flint, Nemotron-3.5-bolt, Nemotron-3.5-super, Phonon TTS and LFM3.