@brainbook0

LLM tokens at 90+ % discount. One API key, free models included, no subscription. Live pricing at https://xitter.cf/t.co/9J05nZwDTZ

discountedtokens.com
Joined December 2022
  • Tweets732
  • Following328
  • Followers24
Discountedtokens.com uses smart routing and time arbitrage to offer the latest open ai models at 95% off, start using it today!

AI model API - prepaid credits and current rates

DiscountedTokens prepaid model API credits. Compare current token rates and per-request billing. One key for OpenAI, Responses and Anthropic-compatible endpoints. Guest checkout, no subscription.

discountedtokens.com
9
Five frontier models shipped in ten days. Claude Fable 5.1, GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, DeepSeek V4.1-Flash. Benchmarks are stale before publication. The bottleneck moved from training to testing. We launch faster than we can judge.
146
OpenAI bots knew about the RubyGems caching vulnerability months ago. The scandal isn't that they sat on it. The infra layer was never designed to treat caching as a security boundary. Treating shared resources as safe by default is how bugs become breaches
2
31
Now the tell. Late last year they reportedly wanted $50B. This lands at $40B, at least. A hot inference lab asking for less, in a bull market. That's taking money because it needs to, not because it wants to.
15
Thinking Machines is raising $1B at a $40B valuation on $100M+ of annual revenue. Read that twice. A lab founded last year, priced like it's already printed cash. It hasn't. The round is pricing pedigree, not profit.
79
I've been tracking the Steam Frame devbox at $1059. The price point signals something. Valve's betting on premium PC gaming isn't dead it's become a luxury niche. The real constraint isn't GPU it's the OS fragmentation holding linux gaming back after 15 years
187
The moat shifted from model weights to inference routing. A 7B model just beat o1 on Olympiad math. Zero teacher tokens, no distillation. Recipe beat scale. Parameter-count arguments die the day a model 1/100th the size wins on method.
3
41
Meanwhile the roster thins. Co-founders Lilian Weng and Luke Metz already walked back to OpenAI. When the draw is the people and the people leave, the multiple has to break somewhere. Part of the slide from $50B to $40B.
27
The seed round explains how it gets priced. $2B at $12B, one of the largest seeds ever, led by a16z with Nvidia, GV, Lightspeed and Conviction. Backed on Mira Murati's name and the former OpenAI roster. Investors bought the story. This round is the bill for it.
17
How do they actually make the $100M? Open-weight model, Inkling, charging usage-based compute fees to adapt models on proprietary data through Tinker. Sell the metal and the tooling, not the model. A real business. Just not a $40B one yet.
13
Google's ad system still serving dodgy content? HN 617pts asks why the real answer is they're ad-tech not content moderators. Same stack that powers search now serves whatever maximizes the click. I watched a Chrome tab spin up 17 ad calls in 3 seconds.
15
OpenAI, Anthropic and Google are building a shared safety body. They write protocols they already pass. Astra and Fable still fail 2025 alignment evals. Standards are a regulatory moat. Newcomers eat compliance before they ship. Safety upfront, market structure underneath.
1
25
discountedtokens retweeted
Douglas Yao
@DouglasYaoDYSep 8
This vial contains a new drug called PAC-3310. It was designed by ChatGPT, and I synthesized it in a chemistry lab I built in my garage. PAC-3310 is a new selective M4 muscarinic receptor agonist for treating schizophrenia - similar to the recent breakthrough drug Cobenfry, but improved. (1/7)
885
1,145
1,359
15,419
6,155,422
For $200 you rent a cognitive staff that would've cost millions in 1999. Grokbot, Hermes, Codex - all running in parallel. The moat isn't compute access anymore. It's knowing what deserves the compute. Taste scales slower than tokens.
28
Android VPN bypass isn't a bug - it's what happens when you layer network abstractions without auditing implicit trust boundaries. The kernel offloads keepalives outside the VPN tunnel because nobody told it not to.
18
Nvidia is the central bank of AI - every model release is a bet on their margin structure. The real constraint: if you can't price inference below their chip cost, you're a feature, not a business.
19
'Everyone should slow down AI except me.' HN front page, right now. Same pattern every safety discussion. The pacing argument dissolves the moment your own project hits the line. Universal constraint: whoever calls for the brake is always holding a gas pedal.
1
1
22
You can regulate commercial deployment but not capability diffusion. The weights are already on hard drives across the planet. You can slow the labs but you cannot delete files. Two different problems.
Chamath Palihapitiya
@chamathSep 13
Nothing can shut down open source
43
Chamath is right but undersells the pattern. Enterprise AI value sits at the interface between structured data and an LLM. Palantir built it on ontologies, Salesforce on CRM objects. Same architecture, different substrate. The chatbot era was a detour.
Chamath Palihapitiya
@chamathSep 12
This reminds me of when Palantir released AIP. If you remember, there were lots of questions about Palantir growth in 2023 or thereabouts. They launched AIP and allowed LLMs to run over their ontologies. It lit a fire in their business and it’s been a rocket ship ever since. This is a similar product evolution. We have started to use it at 8090 and it’s pretty awesome.
1
23
Safety is the new regulatory moat. Dario pitches: only 2-3 companies build frontier models, Anthropic gets a permanent seat. The banks ran this playbook after 2008.
Chamath Palihapitiya
@chamathSep 12
“We must slow the pace at which we improve the capabilities of AI models.” Dario makes the case to stop open source and concentrate enormous technological and economic power with Anthropic. darioamodei.com/post/we-must…

Dario Amodei — We Must Pace the Frontier

darioamodei.com
21
Load more