Aditya Vikram Singh

Ex-VC turned operator, building at the intersection of AI and truth. @IIM_Calcutta and @IITGuwahati alum, CAT'14 100 percentiler, NTSE, KVPY, JSTS Rank 1.

[Aditya Vikram Singh @adityavsingh]

The Universe Is Not a Machine. It’s a Mirror.

Dark sector as hidden awareness.
Non-dual traditions as convergent priors.
The cosmos dreaming itself awake.

Read if you’re ready to stare at the 95% we can’t see.

https://t.co/jLngCl9rF0

[Aditya Vikram Singh @adityavsingh]

@recap_david CONTEXT

[Aditya Vikram Singh @adityavsingh]

Grab yours. https://t.co/AHrquyz0ru

[Aditya Vikram Singh @adityavsingh]

Let's go! @0xSero @localdotai

[Aditya Vikram Singh @adityavsingh]

CodeBurn connects spend across Claude Code, Codex, Cursor, Gemini + others to projects, sessions, retries, commits and PRs.

The question we want to answer next:

How much useful work are you getting per dollar of AI spend, and how do we improve it?

[Aditya Vikram Singh @adityavsingh]

AI coding has a weird accounting problem:

You might know exactly how many tokens you burned.

You usually can’t tell you what those tokens actually produced for you in terms of usable work.

Today @TorukMakto1406 and I are launching CodeBurn, true local-first visibility into AI coding spend, work and waste.

9k+ GitHub stars. ~170k downloads. Free + open source.

https://t.co/2JYr4AkE4h

[CuiMao @CuiMao]

大家好，我们即将迎来Seedance2.5，请允许我榨干Seedance2.0所有性能配合最新Kimi K3模型为这个人类历史上具有里程碑意义的视频模型画上一个完美的句号。【完整游戏流程请移步至评论区点击链接体验】

[Eidzoku @eidzoku]

Codex might be seriously broken right now.

Yup, Codex rant part 3

GPT-5.6 Sol is burning through quota at a completely stupid rate and everyone notices it.

Here’s my usage graph:

Jun 23, GPT-5.5 xhigh: 327.8M tokens
Jul 2, GPT-5.5 xhigh: 549.9M
Jul 20, GPT-5.6 Sol High: 3.3B
I had a bunch of resets, sure, and several tasks were running. That let me burn more.
It doesn’t explain how 5.6 chewed through 3.3B tokens in one day. I genuinely don’t know what the fuck I would’ve had to do to get 5.5 anywhere near that.

And apparently it’s not just me.

@theo already covered 3 different ways GPT-5.6 was nuking usage. Looks like there might be a fourth.

5.6 uses the new Code Mode tool path. It can batch independent tool calls with Promise.all, but it barely ever does. Instead, it keeps running them one by one.

That means another model cycle for every call, dragging the giant context along with it every single time.

One trace found Promise.all in only 5 of 739 GPT-5.6 exec cells:

https://t.co/IQ6gqGWhSK

My own retained logs show the same shape. After removing fork-history replay, the 5.6 peak had 23,791 model requests. My clean 5.5 baseline had 3,803.

That’s 6.26x more requests, while tokens per request were actually slightly lower.

It wasn’t writing much bigger responses. It was just going back to the model over and over and over again.

Then someone tested explicit batching across two unrelated codebases. In the repeated High/XHigh runs, it cut model cycles by ~52–55% and weighted usage by 27–45%:

https://t.co/FGvWZqE4Xm

I’m not saying this one thing explains every token I burned. Several tasks were active, my local rollout logs aren’t the billing ledger, and we still don’t know how much of the spike came from agents themselves.

But both issues are still open, and this looks pretty fucking real.

For now, I’m testing explicit batching guidance. I’m also avoiding the built-in subagent/fork workflow for long-running work and moving bounded jobs into fresh disposable Codex tasks instead.

Self-contained prompt, fire and forget, inspect the result in the main task, archive it. Less context coupling, and each job is actually measurable.

I hope other people can validate it, too.

@thsottiaux @reach_vb @OpenAIDevs if it's true, it should be addressed. Cuz it might be THE cause.

[Aditya Vikram Singh @adityavsingh]

@igorimx @willdepue You can just tell it to reuse the same session, it does that.

[OmedTheVibeCoder @OmedVibeCodes]

OpenAI’s limits are an absolute disaster right now.

I’m on the €200 ChatGPT plan, and my limits are gone after just two days. With GPT-5.5, I could work for seven days straight without hitting them once.

Claude’s limits feel easily 10x better, and Opus 4.8 is honestly better than GPT-5.6 Sol for me anyway.

I genuinely don’t understand why so many people are still using GPT-5.6. OpenAI has never had limits this bad.

[Tyler @rezoundous]

dare I say, Claude now has better limits than Codex

[Aditya Vikram Singh @adityavsingh]

@thsottiaux Please fix the limits! I was just praising these publicly 2 weeks back.
Sol Medium takes 13% weekly quota per hour now. On a 20x plan!

[Aditya Vikram Singh @adityavsingh]

@jturntdev Okay, this is a joke now. 13% getting consumed PER HOUR. on Sol Medium.

[Aditya Vikram Singh @adityavsingh]

@jturntdev How is one supposed to work like this? This is when I reset it 9 hours ago, when the previous weekly session ran out in 2.5 days. Using nothing higher than Sol Medium.

[Toruk Makto @TorukMakto1406]

CodeBurn Desktop is here.

The full CodeBurn experience as a native app: the same local-first engine behind the CLI and menubar, now with the room a real dashboard deserves.

- Overview: 30-day spend, daily activity, and an efficiency score in one screen
- Sessions: all of them, priced and ranked
- Spend: cost flow from model to project, day by day
- Optimize: findings that surface recoverable waste, retries, and redundant context
- Compare: two models head to head on your real work, not on benchmarks

As always: everything runs locally, nothing leaves the machine, open source, free.

Available for macOS, Linux, and Windows.

https://t.co/iU5ZC1qYXj

[Aditya Vikram Singh @adityavsingh]

There's a bunch of memory systems for agents floating around, tried a number of them, but nothing quite hits the mark. What are folks actually using, and what's working?

[Codeburn @_codeburn]

CodeBurn Desktop is here.

The full CodeBurn experience as a native app: the same local-first engine behind the CLI and menubar, now with the room a real dashboard deserves.

- Overview: 30-day spend, daily activity, and an efficiency score in one screen
- Sessions: all of them, priced and ranked
- Spend: cost flow from model to project, day by day
- Optimize: findings that surface recoverable waste, retries, and redundant context
- Compare: two models head to head on your real work, not on benchmarks

As always: everything runs locally, nothing leaves the machine, open source, free.

Available today for macOS, Linux, and Windows. Link in the first comment.

[Aditya Vikram Singh @adityavsingh]

Codex ftw, 5.6 finds edge cases even fable missed, noticed fable repeatedly taking fair pushback from 5.6 sol high. Plus limits!

[Aditya Vikram Singh @adityavsingh]

@RickDavis404 @TorukMakto1406 @thsottiaux @OpenAI @reach_vb Vaibhav, can you help this guy? OSS 6-month plan granted, claiming the plan got his account frozen. Seems weird.

[Aditya Vikram Singh @adityavsingh]

@bryan_johnson You'll be the first, @bryan_johnson.

[Aditya Vikram Singh @adityavsingh]

@hyojun_at @AsideAI 800 subagents on 11 layers, man.