AI tool prices rarely rise with an announcement. They change in the fine print.
So I built https://t.co/zg1WuIjzI6. It checks pricing pages daily and logs every change, dated and sourced.
AI assistants, notetakers, agents, video and voice, email.
Free weekly email.
Google just sweetened its $99.99 AI plan with something that isn't AI.
Google AI Ultra now includes YouTube Premium Family: ad-free YouTube for you and up to 5 people in your household. Yesterday it was the individual plan.
YouTube sells Family on its own for $26.99 a month.
Smart bundle. It also makes it harder to tell what the AI part actually costs.
I track Google's AI plans and what's in them at https://t.co/51cTBjrLXw
Is that an AI subscription or a household bill?
#MadeWithAI #Gemini #AIPricing
OpenAI just put a price on patience.
New in its API: tell GPT-6.2 you can wait 2 hours for an answer and pay 30% less per token. Wait 8 hours, 40% less.
Good fit for overnight work. Meeting summaries, document review, the morning's support tickets.
Fine print: OpenAI's docs say new models starting with GPT-6.2 don't get Flex, the half-price tier older models still have. The Batch API still lists GPT-6.2 Sol at half price.
The discounts are real, but they only reach you if the tools you pay for are built to wait.
Would your business wait 8 hours for a cheaper answer?
#MadeWithAI #OpenAI #AIPricing
@farzyness A chip that runs one model only pays off if that model's inference volume is huge and cost per token keeps dropping.
Jevons Paradox in silicon.
Small businesses win if the savings show up in plan prices and API rates, not just in margins.
@BrianRoemmele Nice!
Thousands of apps is exactly when curation matters.
Do you flag which ones need account access or can send or buy on your behalf?
That's the first filter I'd want before plugging any of them into my Grok Bot.
The benchmarks get the headlines, but the price is the real story.
Neon is $3 in / $15 out per 1M tokens on Google's API page, with thinking tokens billed as output.
Frontier-level at that price changes the math for small shops building agents.
Jevons says usage goes way up!
Day-one pop or not, the harder question for $ANT holders is pricing power.
Google just dropped a frontier model at $3 per million input tokens.
When the alternatives keep getting cheaper, how long do premium API prices hold?
Follow the money: customers go where the price is.
@codyschneider Saving this for every small business owner who's getting pitched a new "AI visibility" retainer.
If the assistants are reading search results, the work is the same old work: answer real questions clearly and get cited by sites people trust.
New acronym, same invoice.
@elonmusk Running a few for my own business right now.
Best advice I can give: one bot, one job, and you review the work like you would a new hire.
For a small business, the cost of entry is the $30/mo Premium plan.
Cheapest employee you'll ever have, but they still need a manager.
@MarcelaMarjean To quote Ronnie, “Trust but verify”
Every sane person should have at least a few days worth of supplies - ‘just in case’.
More if you can afford it.
@farzyness 100%. The decoder ring has to include price, not just quality.
Model prices move every week.
If the router picks the cheapest model that's good enough for each task, and shows you why, small businesses will never go back to paying flagship rates for grunt work.
Anthropic starts trading today as ANT.
Last night it priced at $83 a share, above its $70-$78 range. $81.3 billion raised, the biggest IPO on record. About $1.5 trillion fully diluted.
What matters if you pay for Claude: it now reports to shareholders every quarter.
Claude Pro is $20 a month this morning, $17 on annual. Anthropic's own filing says it expects it may change its pricing models.
I check Claude's plan prices every day at https://t.co/vA8GyvgmHT
Does Wall Street make your AI cheaper or pricier?
#MadeWithAI #Anthropic #Claude
To the broadcast and video production nerds out there...
This post came out a few days ago and I can't get it out of my head.
Look up 'Prime 360' this sounds like 'bullet time' from the Matrix movies on steroids!
https://t.co/ghW1qsMuft
Amazon is rebuilding NBA plays in 3D with AI. During live games.
This isn't a drone. It isn't a video game. It's real game footage from a camera angle that never existed.
It's called Prime 360. Here's how it actually works:
1. A ring of cameras
High-res cameras sit in a full perimeter around the court. Every one is calibrated and synchronized, catching the play from all angles.
2. The play becomes a 3D scene
The footage goes to the AWS cloud, where the play is reconstructed as a 3D scene. Once it's a scene, the camera is free. You can fly it through the play, down on the floor, behind a shooter, right next to the ball.
3. Diffusion models paint the picture
Amazon says it runs on diffusion models, the same class of AI behind image generators. Each frame gets split into dozens of overlapping tiles that go through the model one by one.
4. Custom chips make it fast enough for live TV
That's billions of computations per frame. Amazon runs it on its own Trainium chips and says it cut latency by more than 40%, enough to use during live games.
What people miss:
This launched in October. It blew up Friday night because a Knicks vs Pacers reel had everyone saying it looks like NBA 2K.
It's not on every game. Prime 360 shows up in select NBA on Prime broadcasts, in-game and in pregame, halftime and postgame breakdowns.
And Amazon hasn't said how many cameras are in the ring or how long a replay takes to build.
Why it matters:
For years we asked when AI video would look real. This goes the other way. Real footage, rebuilt by AI, so a camera can go where no camera can.
Sports broadcasts are starting to look like video games. Soon video games will look like broadcasts.
Would you watch a whole game like this?
@minchoi Good timing on this.
Anthropic lists on Nasdaq today.
If their own team runs dozens of Claudes in parallel, the next question for the rest of us is what that costs per developer, and whether the price of Claude Code stays where it is now that there are shareholders.
Rate cards: Neon is $3 / $15 per million tokens on Google's page, Opus 5.5 is $4 / $20 on Anthropic's.
So the list price is 25% lower.
The rest of the 'half' has to come from fewer tokens per task.
That's the number that matters, but it's also the one that moves the most from one job to the next.
Alibaba just priced its new flagship, Qwen4-Max, at $1.50 per million input tokens and $4.50 per million output.
On Anthropic's own pricing page:
Claude Opus 5.5: $4 in / $20 out
Claude Sonnet 5.5: $2 in / $10 out
The API price war isn't slowing down.
But price per token isn't price per job. A model that burns more tokens or needs a second try can cost more by the end of the task. Even Alibaba sells this on cost per task, from its own benchmark.
A cheaper model only saves you money where you pick the model. Open-source agents like OpenClaw and Hermes let you bring your own key: https://t.co/BqNR2EoNDl
Would you run a Chinese lab's model in your business to cut the bill?
#MadeWithAI #Qwen #AIagents
@BrianRoemmele Enterprises at least have security teams to read this.
A 5-person shop has nobody.
That's why my agents draft and I approve anything they can't undo: no sending, buying, or deleting on their own.
Boring setup, but it's the one that survives a bad prompt injection.