Himanshu Srivastava

Building products at the intersection of AI & software.
Sharing what I learn, build, and discover.

[Himanshu Srivastava @coderhimanshu]

@iamsahaj_xyz I switched to @dia two months ago, and it already has all the awesome features of @arcinternet. Give it a try for a few days, and you might not even miss Arc!

[Himanshu Srivastava @coderhimanshu]

@BunnyCDN After uploading a video, it gets stuck in “Processing” and never seems to finish. The video has been processing for a long time and isn’t being resolved/processed successfully.

Is there an issue with the video processing service right now? Please help!

[Himanshu Srivastava @coderhimanshu]

GPT-6 Astra, build me GPT-7, make no mistakes

[Himanshu Srivastava @coderhimanshu]

astra doing 99.9% on ARC-AGI-3 while sol was at 7.8% is wild

[OpenAI @OpenAI]

This is GPT-6 Astra.

Anything you can do on a computer, Astra can do for you. Fast.

[Himanshu Srivastava @coderhimanshu]

➡️ Coding jumped from 42% to 55.8% in their tests, that's not "incremental" for a .1 release
➡️ Pricing cut is the real sleeper feature, huge if you use it daily
➡️ "Hopefully better" on cybersecurity is doing a lot of work lol, someone's testing that boundary soon

[Claude @claudeai]

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1.

They're the world’s most advanced models for coding and knowledge work.

[Himanshu Srivastava @coderhimanshu]

@mehulmpt makes sense, training distribution always leaks into output style. same reason old code models trained heavy on Stack Overflow used to write everything like a SO answer

[Himanshu Srivastava @coderhimanshu]

People calling AI a "bubble" mostly just want it to slow down before it takes more jobs. Fair feeling, but we already use it too much daily for that to actually happen.

[Himanshu Srivastava @coderhimanshu]

@mehulmpt Couldn't agree more!

[Himanshu Srivastava @coderhimanshu]

@rauchg Curious how this holds up once real tool-calling and state get layered in, that's usually where CLIs creep back up in size and latency. 🤔

[Himanshu Srivastava @coderhimanshu]

Agents actually sending the email instead of just drafting it is the real unlock. The hard part was never writing a good reply, it was trusting AI enough to hit send.

[Himanshu Srivastava @coderhimanshu]

@mehulmpt This already kind of exists, just backwards, tokens/sec is basically just GB/s wearing a nicer outfit. If anything we hide the raw number more over time, not expose it.

[Himanshu Srivastava @coderhimanshu]

@timneutkens Haven't upgraded yet. Migrating this weekend, will report back if it holds on our setup too.

[Himanshu Srivastava @coderhimanshu]

@akshkgd Okay 😅 Here’s the actual project: https://t.co/A9cdb0A0S6

[Himanshu Srivastava @coderhimanshu]

Came across https://t.co/7TlKTnP6xn recently and honestly, this component library feels pretty underrated.

Tried their theme toggle in our ongoing project and the light / dark mode animation is so damn smooth. 🔥

Shoutout to @saurra3h, great work! Definitely worth checking out if you’re building with @shadcn.

[Himanshu Srivastava @coderhimanshu]

Google just dropped Gemini 3 Pro, and the coding benchmarks are WILD.

💻 LiveCodeBench: 2439 Elo
⚡ Beats Claude 4.5, GPT-5.1, everything on the chart.
🤖 Agentic coding: 76.2% on SWE-Bench Verified.

This release is clearly built for DEV power. 🚀👨‍💻

[Himanshu Srivastava @coderhimanshu]

@yashwanth941v Using Arc for split tabs, sidebar, max web view, multiple spaces, themes, and one-tap tab groups.

That combo alone makes it worth it for me.

[Himanshu Srivastava @coderhimanshu]

@PrajwalTomar_ cursor

[Himanshu Srivastava @coderhimanshu]

@DemonSlayerUSA 😍

[Himanshu Srivastava @coderhimanshu]

@theo Is it possible to continue a conversation in a new chat once the thread limit is reached, picking up right where we left off in the previous chat? 🤔