←Ben’s Bites Waybackbensbites.com/p/grok-has-a-chance23 Sep 2025sent to 143,706opened 38.1%clicks 8,468unsubs 172← 18 Sep · 25 Sep → · random

The newsletter for the technically curious. Updates, tool reviews, and lay of the land from an exited founder turned investor and forever tinkerer.


Hey folks, thank you so much for the support - I got a lot of messages and will get back to everyone. I’m hosting a Ben’s Bites x Factory meetup in SF on Oct 8th (after OpenAI Dev Day). RSVP here.115

So, what do we have from the past weekend…

xAI has a new model - Grok 4 Fast128, and this one looks impressive. Eyeballing the benchmarks, it’s roughly as smart as Gemini 2.5 Pro, but at the price of Flash. For prompts <128k tokens, Grok 4 Fast is priced $0.20/$0.50 for input/output with a massive context window of 2M tokens.

I’m genuinely excited to test replacing Gemini 2.5 Flash with Grok 4 Fast for my API use. I’ve never felt that way for Grok. - Keshav

ps: rumours are that Gemini 3.0 Pro and Claude 4.5 Sonnet are launching soon. Maybe this week.

Google Chrome144 is getting a boatload of AI features (US first). That includes:

  • A Gemini sidebar - The current iteration allows you to ask questions with the context of your current tab. Google is promising automatic browsing, a summary from multiple tabs, search over your browsing history, and more coming soon to this sidebar.

  • AI mode in the address bar - Trigger a Perplexity-like search experience instantly.

  • Native scam protections, blocking spam notifications and password safety using Gemini Nano.

Notion is going all in on agents214 (who isn’t?). Their latest iteration of Notion AI is a personal agent that can do everything you can do in Notion, like create databases, search across pages, and execute workflows. Soon you can have more than one—with the coming release of custom agents.

Speed is alpha. Get unlimited access to Brightwave’s powerful AI investment research platform for 14 days. Rapid deal screens, instant memos, and real-time market insights - Brightwave does the grunt work for you. Start your free trial today.71*

*sponsored

🌐 What I’m consuming

  • Next-gen work AI Agents + Assistant functionality, unveiled at Glean:LIVE. Register here146 to watch the virtual launch, which features live product demos, performance results, and new personalisations.*

  • Launch day lies104—day two tells the truth. Naveen (Monologue’s maker) talks about the drop-off after the initial excitement vs a product that people adopt as their daily driver.

  • 6-minute video of adding a heatmap activity feature138 with Droid - Factory’s CLI tool.

  • Why we built the Responses API213. Worth reading if you’re still using the old completions API from OpenAI.

  • From managing people to managing AI307 with Julie Zhuo.

*sponsored

⚙️ Tools and demos

  • Orchids358† - A full-stack engineer with state of the art UI capabilities.

  • X CRM180 - Get a list of people you follow on Twitter in a CRM like table and add notes using MCP.

  • Ray 3 by Luma Labs124 - HDR video generation with draft mode and annotation support.

  • MagicPath Libraries228 - A living doc that AI uses to design in your style, always synced with code.

  • Research a Person on Happenstance432 - Build a detailed profile about anyone.

  • Howie410† - Your personal secretary for managing your calendar like a world-class EA. ($6M seed raised)

  • Ambient461 - A daily briefing for CEOs who want to move fast.

  • Perplexity Email Assistant374 - Personal assistant for scheduling meetings, prioritising emails, and drafting replies for you.

via Perplexity374

🥣 Dev dish

  • tldraw SDK 4.0180 - build infinite canvas apps for the web.

  • Vercel Agent124 for code reviews is now in public beta with $100 in free credits.

  • LLM gateway205 - Use any model, from any provider, with just one API (open source).

  • step.run by Inngest100† - Just wrap any REST API with step.run to add durability, automatic retries, and observability.

My friends at Angel Squad have given me a limited number of 30-day guest passes for BB readers. Their thesis is that product builders make great angel investors (and I agree). TechCrunch calls them the “YC of angel investing” – a 2k+ community giving you access to high-growth startups through Hustle Fund. Focus is early-stage and you can invest as low as $1k, but you’ll also get pre-IPO opportunities like SpaceX, Anthropic, Databricks and Canva.*

*sponsored

💎 Underrated gems

  1. Moondream 3 is a small vision model222 - It is better than GPT-5/Gemini/Claude when it comes to identifying elements in an image – not just simply pinpointing things but complex queries that require reasoning. If you’re building a browser use tool that relies on screenshots, check this out. It might reduce your costs by a big factor while improving accuracy.

Image
via @vikhyatk — creator of moondream32
  1. Scale AI has built a new programming benchmark: SWE-Bench Pro50. The general trend for performance is roughly the same when compared to past benchmarks, but this one measures longer, more complex software engineering tasks with a collection of public and private repos

Leaderboards for SWE-Bench Pro — Public25 and Private408

🍦 Afters


That’s it for today. Feel free to comment and share your thoughts. 👋

📷 thumbnail creds: @keshavatearth41


Grok has a chance
Notion for AI is... Notion

less more clicks · number = clicks † = site gone, link opens the Wayback Machine

Most clicked in this issue

461ambient.us
432happenstance.ai/research
410howie.com
408scale.com/leaderboard/swe_bench_pro_commercial
374×2perplexity.ai/assistant
358orchids.app
307youtube.com/watch?v=c_w0LaFahxk
228x.com/skirano/status/1970187546694144303
222moondream.ai/blog/moondream-3-preview
214notion.com/blog/introducing-notion-3-0
213developers.openai.com/blog/responses-api
205llmgateway.io

3,051 readers clicked at least once. 37 links, 36 with clicks. Heat is relative to the most clicked link in this issue.

Substack · essay