Sabitlenmiş Tweet
The Grid’s Beta is LIVE!
We can get your AI API costs down by up to 80% by making suppliers compete for your requests.
Your first 200M tokens are on us, start building
→ app.thegrid.ai/sign-up
English
The Grid
229 posts
The Grid
@The_GridAI
The spot market for AI Inference
Katılım Eylül 2025
85 Takip Edilen1.5K Takipçiler
Get started by changing the model name and pointing your base URL to api.thegrid.ai/v1.
Live prices for each model are available in the app at thegrid.ai/pricing.
English
Our perhaps unsexy take: a lot of this frustration comes from treating every task like it needs the same frontier "think hard" model. Some work just needs reliable day-to-day coding. Some genuinely needs deep reasoning. Lumping it all into one GPT-5-shaped blob is where a lot of the pain starts.
English
I 1000% agree with the RL-fried notion here. I feel like I’m screaming into the void about this. the models are NOT good at this. I miss the models that actually understood the INTENT. NO amount of prompting can fix this because you spend immense time per prompt writing out every edge case you can think of, only to then have to babysit the internal test time compute of the model thinking to make sure it doesn’t burrow itself in a hole mid-reasoning stream.
and then you spend just as much time writing and fixing your prompts as you would have just implementing, except then you get all the bad aspects of not fully understanding your code unless you take even more time to understand (in which case you are subject to the psychological burnout of understanding the codebase only for it to shift under your feet 5 prompts later, not even including parallel execution of tasks across a codebase)
this is heavily analogous to reward hacking in RL training and a bad exploit/explore ratio. earlier models, namely GPT 4.5 (and Claude still seems to be much much better at this) were trending in the right direction
I sincerely hope GPT 6 fixes this, as all versions of GPT 5 have had this issue in my experience
Josh@jjpcodes
i'm kinda sick of 5.6 sol now honestly. yes it's a good model, no it doesn't help me ship as much software as it should. it feels too RL-fried and ends up in all these local minima where it just spins its wheels doing nothing, invents bullshit overly cautious procedures you didn't ask for, doesn't solve your problems, hyperfocuses on tiny details, adds theatrical bullshit like tests/ledgers/hashing/proofs/review loops to "ensure stuff works" (none of it works) and degrades context over long-running threads. 1/3rd of the time it can distil your intent kinda okay and keep it up for a few hours but it will eventually collapse into total slop cannon, no matter how much you steer it. the other 2/3rd of the time it will just misunderstand you and slop cannon. oh yeah and the limits are cooked and it burns 5x the tokens of 5.5 whilst you don't ship. bring on GPT 6.....
English
@annagrad78 @sama chasing whatever-latest forever can really wear people out, specially when they just want to ship
English
@sama It might be wrong, but we need more stability, and the constant model changes are exhausting. #BringBack4o without guardrails (for really old people), and we wouldn't have to worry about the everlasting updates anymore.
English
@rabahrahil Feels more like a model roulette.
When you pin to one frontier name, every "update" can change tone, assumptions, and how much of the task you actually get back.
English
either i have become AI illiterate in the past two days or the new claude model updates have been horrible.
they are 93% lecture me, tell me x, y and z, wrong assumptions etc and like 7% of the task I asked for.
is it me or are these updates terrible?
English
@Austin_Federa @AnthropicAI Feel like quality really shouldn't feel this brittle between updates/releases.
English
Opus 5 seems like a remarkable downgrade compared to 4.8.
Opus 5 is blatantly lying to me about basic thermodynamics, messing up simple math, and constantly contradicting itself when you ask it to rethink core assumptions.
@AnthropicAI really blew this release
English
@rohanpaul_ai This feels directionally right. I do wonder if part of the problem is paying frontier-model prices for work that never needed frontier-level intelligence in the first place.
English
Boris Cherny, head of Claude Code at Anthropic, on optimizing token cost and model use.
“I use Fable for everything"
There is probably a 50% opportunity to reduce the investment (on token cost). However, there may be a 1,000%, 10,000%, or even 100,000% opportunity to increase the return.
Therefore, I would simply use the most expensive model and focus on how to get more out of it. Ask yourself, ‘How do I increase the return?’ Do not focus primarily on cost-cutting."
----
From "Scale" YouTube channel, (full video link in comment)
English
@MarotinhaM @OpenAI @AnthropicAI But when the bill still feels out of proportion to what you actually got done, that gap can be frustrating at times ...
English
Token smartness is the biggest chalenge we still have.
Look at the numbers of the API cost for my usage in codex and claude code this week. It's doing a great job, but the aggregate value it generated is far from cost.
PS: Had two codex reset this week.
@OpenAI @AnthropicAI
English
@tannishmango @AnthropicAI @cursor_ai Cost surprises like this really sting.
People shouldn't have to reverse-engineer which model is worth it mid-month.
English
Absolute rugpull by @AnthropicAI and @cursor_ai .
Opus 5 is insanely expensive.
Had no clue I was gonna get rekt this hard after Anthropic's social media blitz ab Opus's new "cost efficiency".
GPT-5.6 Sol gets you FAR more for FAR less.
Nuke Anthropic. Destroy Anthropic.
Claude@claudeai
Introducing Claude Opus 5. It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.
English
Sometimes you need your tokens, other times you don't.
The Grid@The_GridAI
What if unused tokens weren't wasted? Reselling is live on The Grid. Use what you need, resell what you don't. Tell your agent to buy and sell inference to stretch your spend.
English
@FutureStacked Every unused token is spend you could've stretched further. Now it doesn't just sit there.
English
@The_GridAI A marketplace for unused tokens just makes sense.
English
@AriaWestcott That's the goal, pay for what you use, get something back for what you don't.
English
@AIHighlight Supply and demand for inference, just like everything else.
English
@TheAIColony Idle compute sitting around unused never made sense to us either. Curious to hear what you think once you've tried it out.
English
@The_GridAI This is actually a smart way to reduce AI costs.
English
@AIFrontliner If you're burning through tokens fast, the waste adds up quick, glad this lands for you.
English
@The_GridAI Definitely solves a pain point for heavy AI users.
English
@The_GridAI "Hey, Hermes, go use The Grid's spot market to optimize your inference spend. Use limit orders to catch the dips, only then run your scheduled tasks, leveraging the discount. Any unused tokens find strategies to resell at higher prices.
Make no mistakes."
English