Cursor not working? Fixes for startup, AI and sign-in errors
Cursor's own help center covers blank screens, blocked HTTP/2, extension conflicts, sign-in domains and slow editors. Here are the fixes in the order to try them.
770 stories
Cursor's own help center covers blank screens, blocked HTTP/2, extension conflicts, sign-in domains and slow editors. Here are the fixes in the order to try them.
Install steps from the official docs, plus the memory each Gemma 4 size needs and what to check when a model won't load.

Google released Gemma 4 under the Apache 2.0 license, so you can download and use it commercially at no cost. Hosted use and hardware are separate.
The prototype carries four TPUs and runs one at a time for about 20 minutes. Google plans two more satellites in 2027.
The webcast is at 5 p.m. ET. Last quarter revenue was $1.935 billion, and Palantir guided the third quarter to about $2.16 billion.
The open-weight model built working attacks on six Firefox releases from public fixes before users received them, in up to 48 hours per attempt.
Reddit beat on revenue at $895M, but said search referrals are dropping and daily users could fall by a low-double-digit percentage in Q4.

Apple's HomePod mini 2 goes on sale October 30 without Siri AI. Amazon's new Echo Dot is $49.01 cheaper and ships with Alexa+, which Prime includes.
The model cards give no hardware list. The vLLM and SGLang recipes start at eight H200 GPUs for GLM 5.3, and Flash is smaller but still a server job.
AMD's call is at 5 p.m. ET. Last quarter revenue was $11.5 billion, and AMD guided the third quarter to about $13 billion.
Z.ai put GLM-5.3 and GLM-5.3-Flash on Hugging Face. Flash is MIT, the full model has a custom license, and API prices start at $0.15 per million input tokens.
Cursor says a modified build in its marketplace on October 29 tried to plant an SSH key and a reverse tunnel, affecting at most 5,000 users.
A new training method scores 94.25% on Chinese long-text rendering at two network evaluations, with small gains elsewhere and one benchmark it loses.
Call it as mistral-large-4 on Mistral's API, use the demos while the free budget lasts, or serve the open weights with vLLM on eight GPUs.
Anthropic's specs show the two models match on context, output and cutoff. Opus costs twice as much and is slower, and Sonnet cannot turn thinking off.
Anthropic's pricing page shows Opus as unavailable on the Free plan. Pro costs $20 a month, and the API charges $4 per million input tokens.
Serve Qwen3.8-27B with vLLM on one or two RTX 5090 cards, or use a GGUF quantization, with the settings Alibaba recommends.
The 27B model is Apache 2.0. The 2.4T and Flash-Next weights carry custom Qwen licenses, and Max and Flash are paid API models from $2 and $0.15 per million input tokens.
Gemini starts at $0 and sells AI as its own plan. Copilot comes bundled in Microsoft 365 from $9.99 a month.
Strava subscribers can link their account to Gemini, read-only, and ask about pace, races and trends. It is US-only at launch.

A NeurIPS 2026 paper finds 10 of 25 popular benchmarks have a lopsided difficulty mix and 8 leave wide gaps, then proposes BenchFill to patch them.
Where to see your Codex limits and reset times in a ChatGPT plan, in the browser and in the command line.
Claude is a general assistant, Perplexity is a cited answer engine. Both start free, and Pro costs $17 to $20 a month on each.
OpenAI's GPT-6 models score 55 to 57 on Artificial Analysis against 32 for DeepSeek V4.1 Flash. Prices and scores as of October 30, 2026.