Shane Wray

MS student specializing in monetary economics, tech enthusiast, hotkey master

[Shane Wray @Shane_Wray]

Where’s the Codex quota reset market @Kalshi @Polymarket?

[Matthew Berman @MatthewBerman]

Codex Reset prediction: 26%

Keep conserving quota, brothers

https://t.co/oJDOzAMNSX

[Shane Wray @Shane_Wray]

@Conor_D_Dart ...or it expires in 30 mins. Tibo's telegraphing the resets is helpful

[Shane Wray @Shane_Wray]

@AdamHoltererer Let's be honest, your quota gonna 2x and your wallet -$100

[Shane Wray @Shane_Wray]

@2sush How many of those "port old games to Mac" were just it installing Wine?

[Derek Nee @DerekNee]

turns out a lot of you noticed the same thing.

don't expect an official response, so i had my agents build something:

https://t.co/VypF24GVBy

→ independent benchmarks for codex & claude code every 2 days
→ problems sourced from TerminalBench2 (easy ones filtered out)
→ subscribe (free) to get alerted when something gets nerfed

first batch is already running. scores drop in 2 days.

if this gets 3k subscribers i'll keep it running long-term (benchmarks burn a lot of tokens and that's not free).

RT appreciated. let's keep them honest.

[Shane Wray @Shane_Wray]

@thsottiaux Unfortunate re 5.3 Codex—it was useful for efficient subagents.

[Shane Wray @Shane_Wray]

@Soso_fun_yt @OpenAI It'll go back to normal on May 26 when your week refreshes.

[Shane Wray @Shane_Wray]

@Shinnzo_xd @Soso_fun_yt @OpenAI False.

[Shane Wray @Shane_Wray]

@RussekFilm @HistoryBoomer Nonsense. Unless you're in an expensive city, get all your veggies fresh, and don't use Amazon S&S to buy in bulk, anything above $10/day is a luxury e.g. choosing organic for everything. I eat very well (and almost all organic) and average $8.60/d

[Shane Wray @Shane_Wray]

@lydiahallie So, the programmatic use credit is not useable for "extra usage" in CC?

[Shane Wray @Shane_Wray]

@RichardHanania Too easy. Sonnet/Opus 4.5-4.7 and GPT 5.2-5 each have tells that were trivial to spot. The real test is to load up agents with your writing samples and style guides to remove their tells (e.g. Opus using space-emdash-space, GPT being autistic)

[Shane Wray @Shane_Wray]

@KaiXCreator This needs a community note

[Shane Wray @Shane_Wray]

@hiarun02 "Codex usage limits are shared with other agentic features. This currently includes Codex and ChatGPT for Excel."
Doesn't seem to include ChatGPT web

[Shane Wray @Shane_Wray]

@andrew_n_carr Banger. System prompt:

[Shane Wray @Shane_Wray]

@sakpo0007_ Interesting how none of these answers are right (historically). Paul, like Jesus (e.g. Matthew 16:28, Mark 9:1, Luke 9:27), thought the end of the world is coming soon—in their lifetime. Just look at the next sentence: 1 Cor 7:29-31. Also 1 Thes 4:13-18 etc.

[Shane Wray @Shane_Wray]

Can you try fixing the "Agent terminated due to error" first? That'll help the reduce load from users mashing the retry button. Also, I'd love to use Flash more, but I don't enjoy it asking to run a terminal command just to write broken text to a simple .md in the project folder "T�h�i�s� �c�o�l�l�e�c�t�i�o�n�..."

[Shane Wray @Shane_Wray]

I really appreciate that @OpenAI's /fast and 1m context modes can use the subscription quota rather than requiring extra billing on @claudeai

[Shane Wray @Shane_Wray]

@SawyerMerritt Pretty disappointing. Yes, 8gb is low, but the real bummer is the weight: 2.7lb is same as MBA

[Shane Wray @Shane_Wray]

@rstormsf All true. You also have to be guide it around editing its own configs—Sonnet and K2.5 made malformed edits requiring doctor --fix, and I had to make Codex ssh in once to fix what doctor couldn't

[Shane Wray @Shane_Wray]

@ryancarson @openclaw Yes, but there are many cases where you shouldn't have the main session run routine ops—heartbeat carries the session context and can be token intensive. I find isolated/subagents better for most scheduled jobs