Pinned Tweet
Inspired by the incredible real-time AI vision demos and ngxson's live SmolVLM + llama.cpp project, I wanted to see what was possible on Apple Silicon! 🍏
Super excited to share my SmolVLM Real-Time Webcam Demo with MLX-VLM! 😙
Thanks aimanamri and neverblinks for contributing and sharing! The Claude Code desktop themes look awesome!
buildwithclaude.com/themes
20
Gemini 4 Neon has rolled out in Google Antigravity. Interestingly, Google also offers an Orchestrator (Neon) model designed primarily to coordinate and delegate tasks to Hydrogen.
Gemini 4 Neon is rolling out in Google Antigravity!
Bringing frontier-level intelligence and ranking #1 on the Artificial Analysis Coding Agent Index, Neon is rolling out today across the desktop app, CLI, and IDE Extensions for paid Google AI Pro and Ultra subscribers, and for enterprise users through Gemini for business.
You can also use Orchestrator to pair Neon as the planner with Hydrogen as the executor, giving you Neon's judgment where it matters and Hydrogen's speed everywhere else.
2
2
45
Snow that falls harder as your context fills.
If you want one today, Snowfall by Brezn610 does exactly this, and it's now on Build with Claude with install commands:
buildwithclaude.com/themes?q…
Custom themes are just mods so your background can even react to your actual session!
Here I just asked Claude to make it snow as my context fills up (the mod hooks into turn.step) but you can get really creative with it ☃️
39
Custom themes for Claude Code desktop just landed, so I added a Themes section to "Build with Claude".
59 themes so far across desktop and terminal, each with a preview and install commands. All 20 official scenes are included.
Built a theme? Submit a PR. Get it seen by an audience of 100K+ monthly viewers.
buildwithclaude.com/themes
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own.
Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
294
Google launched a Decisions API this week, so it went into my Trust Me Bro Phishing test with Jev and the other Decisions models 😄
200 real phishing emails, plus 200 normal emails written to look like them. Same questions and rules for every model.
Phishing caught / false alarms:
- TypeSafe Jev: 180 / 8
- OpenAI Decisions API: 185 / 3
- Cloudflare Clef: 191 / 7
- Gemini Decisions API: 188 / 1
1 false alarm in 200 is the lowest of any AI model I've tested. It beat Jev on both counts, and the gap is too big to be luck. Clef still catches a few more, but that gap is too small to call.
Downsides: about 2x Jev's price, and 709ms per email from Sydney. Around 350ms of that is just the trip to Google's servers.
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
- 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring
- $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)
- leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
2
205
Wow, the performance and calibration are insane! And it supports video too?!
I definitely need to run another phishing email test with this model and the other decision models to see which one actually performs better!
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
- 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring
- $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)
- leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
30
Another one
Security release pre-alert:
We will release new versions of Node.js 22, 24, and 26 release lines on or shortly, after 2026-11-11 in order to address security vulnerabilities.
The highest severity issue fixed in this release is HIGH.
nodejs.org/en/blog/vulnerabi…
27
Anthropic rewrote its Messages API in Rust. Zero Rust devs on the team.
Claude wrote the code. Humans wrote the plan and checked old vs new matched on real traffic. CPU dropped 40-50%.
Pretty sure "never do the big rewrite" is dead advice.
@katelyn_lesse on @latentspacepod
The $30 Premium plan seems like pretty good value. You get access to all the Grok models, Grokbot, and 4x more usage.
DeepSeek V4.1 Flash is also available in Cursor now, and it's super cheap to run, so you can get heaps of tasks done at a low cost.
Might be worth checking out 😁
DeepSeek V4.1 Flash is now available in Cursor!
Every effort level sits on the CursorBench cost frontier: from $0.60 to $0.98 per task, no model scores higher.
Turn it on in Cursor Settings > Models.
2
138
Anthropic is reportedly going public in a few weeks, so there are a few reasons these people might be leaving. Either they have all their shares and can sell them after the IPO because Anthropic messed up their contracts, or Anthropic fired people to save shares, or something really bad happened. 🤔
73





















