What an insane day in AI. The frontier models just became substantially cheaper, with the Opus 5.5 price cuts, and now with GPT-6 Sol and Luna dropping token prices by 50%. The rate at which the cost per task (on a like-for-like basis) drops in AI is unlike any other type of technology in history. And every time the cost of AI drops, the use-cases you can deploy agents against dramatically increase. This is Jevons paradox applied to agents. These improvements will directly lead to broader diffusion of AI in the economy as we can use agents to process all of our data, scan our code for security issues, read through all log data to make decisions, have agent swarms in workflows, and much more. The cost of tokens is directly correlated to these use-cases being opened up at scale.
OpenAI
@OpenAI
Sep 22
Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe. GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale. We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.

Sep 22, 2026 · 7:16 PM UTC

108
84
748
142,755
Sort replies: Relevant Recent Liked
Replying to @levie
cheaper models make more workflows worth trying – agent economics are shifting fast
anish karan
@_anishkaran
Sep 22
The obvious is missing. So we built Sol - hellosol.app Sol finds the work itself, does it, and comes back for your approval. Every day in our emails we say "I’ll share”, "I'll review”, "I'll get back" - then repeat the exact same thing to an AI. Why? Sol finds everything you said you’d do & gets them started for you. It does the research, creates the doc, builds the slides, finds the time, connects the dots across multiple emails, doing everything it takes to get the job done - but doesn’t send, schedule, or share anything until you approve. Sol runs on its own computer, uses a browser, and has a library of skills that automatically get assigned to the work that needs to get done. No setup. It just starts working. We've raised $4M from General Catalyst, Nexus Venture Partners, DeVC, PeerCheque, Kunal Shah, and a few others. Extending early access now. @generalcatalyst @nexusvp @DeVC_Global @peercheque @neerajarora @b_jishnu @kunalb11 @miten @RTinkslinger @Rahul_J_Mathur @AkarshS27 @SiddhantD06 @RajatAgarwal167
1
90
Replying to @levie
my daughter showed me these and said i should use them because they are "cute af" whatever that means hahaha
59
Replying to @levie
Jevons paradox means cheaper AI will lead to more AI use. So how do we tell whether we're making meaningful progress, rather than just consuming more? Is cost per successfully completed task the right measure? If so, have you figured out a way to measure it.
1
31
Replying to @levie
the cheaper high quality models get the more use cases unlock to increase the adoption flywheel
1
14
Replying to @levie
I can't wait to see what it does for my stockpile of Adderall & Xanax
4
2
773
Replying to @levie
Competition on commodity problems drive price to their marginal cost to serve. Yay economics!!
4
418
Replying to @levie
pixel mascots for models that read your logs and flag security holes. sure
2
132
Replying to @levie
Like-for-like cost per task is the right unit for that claim. A 50% list-price cut only opens new use cases if tokens per finished run hold. When a model is cheaper per million tokens but burns more output to reach the same quality, the diffusion curve moves slower than the price sheet suggests.
1
149
Replying to @levie
I spend around $400 a month on Claude, still waiting for that number to go down
1
1
433
Replying to @levie
every price cut I go "great, I'll save money" and then immediately spend it on 4x more agent runs. token savings have never once made it to my bank account
1
19
Replying to @levie
Every time token prices plummet by half someone somewhere discovers another internal spreadsheet that can be replaced by an autonomous agent swarm. In our agency where ninety percent of operations run on AI the real bottleneck is no longer compute cost but workflow orchestration and deterministic sanity checks. When compute becomes virtually free what becomes the rarest bottleneck in tech?
1
87
Replying to @levie
cheaper tokens = i can run my ads agent on every client ad instead of just the big spenders. that's the real unlock for small agencies
1
37
Replying to @levie
Large Enterprise is still moving way too slow, the bottlenecks are people and legal/compliance imo but mostly just people not able to keep up to date fast enough
2
138
Replying to @levie
I'd like to see this used on the shared drive nobody wants to open. Finding the old price sheet that still gets sent to customers is a fairly unglamorous reason to process a lot of documents.
1
113
Replying to @levie
Cheaper tokens don’t just cut costs. They expand what’s worth automating. As intelligence gets cheaper, orchestration becomes the bottleneck: where agents run, what context they get, and how their work is verified.
1
109
Replying to @levie
The Jevons paradox framing is the exact economic lens people miss. Dropping inference costs doesn't shrink the TAM. it turns speculative, low-margin agent pipelines into viable enterprise ROI overnight.
1
102
Replying to @levie
cost per task dropping is one thing, cost per correct task is the number nobody posts
1
137
Replying to @levie
Jevons needs the buyer to keep the budget, Harvey answered the token bill by moving to Kimi K3, cheaper can also mean elsewhere
1
87
Replying to @levie
I'd love to see this make voice reminders cheap enough to tuck into every app.
1
118
Replying to @levie
wonder which internal workflows stop being too expensive first
1
50
Replying to @levie
Wild day in AI. When frontier level reasoning gets 50% cheaper, we make entire categories of work economically viable that weren’t worth doing before. The unit economics of intelligence may end up being the real breakthrough.👆
1
32
Replying to @levie
The Jevons part I keep coming back to: cheaper reasoning doesn't buy you cheaper outcomes, it buys you more decisions you can afford to make. And the bottleneck moves with it. When a task costs dollars, you ration generation. When it costs cents, you ration judgment, because every cheap decision still needs someone to own whether it was the right one. The scarce input stops being tokens and starts being people who can tell a good decision from a bad one.
1
233
Replying to @levie
cheaper tokens make the boring jobs worth doing. like running every scraped post through a classifier instead of sampling a few. which use case do you think opens up first?
1
26
Replying to @levie
US frontier labs have no choice. They have to increase efficiency and decrease cost. Else models like deepseek, GLM, MiniMax, Kimi and co. Will replace them very fast.
1
72
Replying to @levie
Yes, but how much of our problems are actually intelligence-bound?
4
2
610
Replying to @levie
The bottleneck will be people. It's already happening in businesses that are implementing AI. You can only implement as fast as people can absorb the information. No one is going to allow AI to make decisions without human oversight (at least for the next several years).
1
42
Replying to @levie
toly 🇺🇸
@toly
Jun 26
.@grok your next models better be called Burnie boden and pnut
1
47
Replying to @levie
Cheaper tokens only widen agent use cases when every successful task still attributes model, tokens, fallback, and outcome. Price cuts without a trail still hide waste. modelbeat.ai keeps that cost-per-task layer measurable. #ModelBeat #Elytra

ModelBeat · LLM gateway and model router for AI products

ModelBeat is an OpenAI-compatible LLM gateway: one base URL in front of every model you are approved to use. It picks the model per request, absorbs the failures, prices the call, and never moves...

modelbeat.ai
8
Replying to @levie
Jevons Paradox is exactly right. But the bottleneck is shifting from token cost to organizational adoption. Cheap agents mean nothing if the enterprise can't trust them with real data. The winners will be the ones who solve trust, not just cost.
20
Replying to @levie
The second-order effect is what matters in medicine. At falling cost per task you can afford to have a model read every report, every trial and every guideline for every single patient, not only the hard cases. That turns AI from a consult into infrastructure.
1
26
Replying to @levie
cost per task dropping this fast is wild. curious which workflows become viable now that weren't 6 months ago?
16
Replying to @levie
In healthcare a licensed name still eats the liability when AI output touches a patient. Token prices halving doesn't touch that rate card. The token was never the expensive part. Who prices the signature?
1
59
Replying to @levie
Agreed. github.com/openai/codex-secu…

GitHub - openai/codex-security: OpenAI's Codex Security CLI and TypeScript SDK for finding,...

OpenAI's Codex Security CLI and TypeScript SDK for finding, validating, and fixing security vulnerabilities. npm: https://www.npmjs.com/package/@openai/codex-security - openai/codex-security

github.com
83
Replying to @levie
Went to sleep on 5.5, woke up to Sol 6 and two context resets. Wild morning.
105
Replying to @levie
jevons only pays the seller if volume outruns the price cuts. the late rounds i look at price both at once, usage triples & the revenue line still compounds. which of those two breaks first if one has to?
32
Replying to @levie
Cheaper tokens can make continuous analysis and multi agent workflows economically practical. Recalculate cost per completed task, then test the use cases that were previously too expensive to run.
1
29
Load more