Which chat tool are you using there to run the model with a Google search tool?
@simonw
-
Claude Sonnet 4 1M Context Window Access Requirements
By
–
Notes on the new 1m context window for Claude Sonnet 4: https://
simonwillison.net/2025/Aug/12/cl
aude-sonnet-4-1m/
… You need to send a beta header of context-1m-2025-08-07 and be on tier 4, which means you have purchased at least $400 in API credits -
Property Deeds Blockchain Security Risks and Phishing Vulnerabilities
By
–
That property deeds thing remains an incredibly bad idea If someone phishes you do they get to keep your house?
-
Basic LLM Chat Terminal Interface Development
By
–
There's an "llm chat" basic terminal chat interface, nothing fancier than that yet though
-
Project-specific feature enablement without account contamination
By
–
Can we have a version of this that can be persistently turned on for specific projects without infecting the rest of our accounts?
-
LLM Command-Line Tool Releases GPT-5 Support Template Features
By
–
New release of my LLM command-line tool and Python library for interacting with Large Language Models – includes support for the GPT-5 model family and improvements to how tool calling works – you can now save multiple tool configurations in a template!
-
Running Qwen Image Model on Mac with 60GB RAM
By
–
Ivan's qwen-image-mps Python script is excellent – if you have 60GB of RAM to spare (somehow my 64GB M2 just managed to run it) you can generate images using Qwen's Qwen-Image model running on a Mac – my notes here: https://
simonwillison.net/2025/Aug/11/qw
en-image-mps/
… -
Thinking Models Show Novel Behavior Patterns in Token Usage
By
–
I've never spotted a thinking model pushing back like this at all before – they usually burn a bunch of tokens planning out which components they're going to draw and where they're going to put them
-
AI Model Demonstrates Responsible Behavior in Output Generation
By
–
This model is pretty sassy, later in the thinking trace it said: Self-check: Am I being too pedantic? Nah—if someone asks for impossible things, it’s better to gently correct than make fake art that could confuse them.
-
Qwen3-4B First Model to Reject Unrealistic Test Requests
By
–
Qwen3-4B-Thinking (a 4GB download) just became the first model I've tested to directly push back against my pelican riding a bicycle test, calling it "oddly specific and completely unrealistic" and saying "this request violates physics and biology" https://
simonwillison.net/2025/Aug/10/qw
en3-4b/
…