Learning Library

← Back to Library

OpenAI Dev Day: Cheaper, Real-Time, Smarter AI

Key Points

  • OpenAI cut its API pricing by 50%, and Anthropic followed suit, marking the first time that AI has simultaneously become cheaper and more powerful.
  • A new real‑time, voice‑to‑voice API (priced around $18 per hour) with token limits up to 10,000 tokens per minute enables developers to build phone‑based automation apps that can rival human labor costs.
  • The company previewed the O1 model, showcasing it live by programming and piloting a drone on stage, illustrating a significant jump in model intelligence and versatility.
  • OpenAI is streamlining fine‑tuning directly on its platform, reducing the need for external middleware providers and allowing developers to create specialized, application‑specific models more easily.

Full Transcript

# OpenAI Dev Day: Cheaper, Real-Time, Smarter AI **Source:** [https://www.youtube.com/watch?v=f5ysxGVEASI](https://www.youtube.com/watch?v=f5ysxGVEASI) **Duration:** 00:06:26 ## Summary - OpenAI cut its API pricing by 50%, and Anthropic followed suit, marking the first time that AI has simultaneously become cheaper and more powerful. - A new real‑time, voice‑to‑voice API (priced around $18 per hour) with token limits up to 10,000 tokens per minute enables developers to build phone‑based automation apps that can rival human labor costs. - The company previewed the O1 model, showcasing it live by programming and piloting a drone on stage, illustrating a significant jump in model intelligence and versatility. - OpenAI is streamlining fine‑tuning directly on its platform, reducing the need for external middleware providers and allowing developers to create specialized, application‑specific models more easily. ## Sections - [00:00:00](https://www.youtube.com/watch?v=f5ysxGVEASI&t=0s) **OpenAI Dev Day Takeaways** - The speaker highlights OpenAI’s recent announcements—including a 50% cut in API pricing, the launch of a real‑time voice‑to‑voice API with high token limits, and the push for massive developer adoption—as signals that AI is becoming both cheaper and more actionable in 2025. ## Full Transcript
0:00open AI had their Dev day yesterday and 0:02I want to call out four takeaways I 0:04thought were significant and I also want 0:05to talk about where this is hinting that 0:07we're going in 20125 so takeaways first 0:10number one they are making intelligence 0:12cheaper they're dropping API costs by 0:1450% on the basis of having seen these 0:17tokens before and so now they can 0:19process them more cheaply and so they're 0:21passing the costs on to the developers 0:24anthropic immediately followed and also 0:26dropped cost 50% this is not a surprised 0:28face over here 0:30intelligence is getting cheaper in fact 0:32we've never lived in a world where 0:33intelligence has gotten cheaper and 0:35better at the same time and that's what 0:37we're living in right now it's a wild 0:39world second thing I want to call out is 0:42we are launching an we are getting a 0:44hold of as developers a real-time API 0:46with voice to voice so this means it's 0:48possible for anyone to build an app that 0:51for example does customer success phone 0:53calls an app that calls up your phone 0:55carrier and demands cheaper rates an app 0:57that calls up the moving company to 0:58arrange a move whatever it is you want 1:00like it's easy to build an app for that 1:02and openai clearly expects a lot of 1:05action from their 3 million plus 1:06developers on that API they're raising 1:08token limits they're going to make it 1:09possible to put 10,000 tokens a minute 1:11through these 1:13apis and they're expecting you to use 1:15the voice one by the hour like they 1:17priced it out so it's a reasonable price 1:18per hour 18 bucks an hour for someone to 1:21talk on the phone it competes with labor 1:25costs and that's exactly where they're 1:26positioning 1:27it okay and that's not the only thing 1:30that they that they did they also 1:31previewed intelligence so I said 1:32intelligence was getting better they 1:34previewed the 01 model by the way I 1:36don't mean 01 preview the model that's 1:38on all all of our apps I mean they 1:40previewed the actual 01 model I know 1:43words are hard so the actual 01 model 1:46they used it to program a drone live on 1:47stage the Drone flew around it shot 1:49video Etc and they did it all like from 1:53setting up the server to Flying the 1:55Drone right there by working with 01 and 1:58that sort of gets me to like so those 2:00are the big takeaways right so we have 2:01the realtime API we have the o1 we have 2:03apis getting cheaper um and they're also 2:05making it easier to fine-tune these 2:07models so like middleware providers that 2:09like lived off of fine-tuning apis may 2:12be somewhat in trouble because they're 2:14going to give you the tools to fine-tune 2:16apis and do smaller application specific 2:18models right off of the open AI sort of 2:21API level so basically you don't have to 2:23go to someone else to tune your model 2:25for you necessarily depending on your 2:27application you can go straight to open 2:28AI for take 2:30but here's the thing like if at the 2:32larger scale the model's getting smarter 2:35with 01 the model's getting cheaper 2:37cheaper tokens affordable voice calling 2:39rates for an hour it's getting more 2:41capable with the fine-tuning with the 2:43voice too realtime 2:45API higher token limits it's pointing 2:48the direction to agent-based workflows 2:51that's the takeaway for me like we are 2:53seeing the shape of what agent-based 2:54workflows are going to look like and 2:56that's actually what Sam also hinted 2:58when he talked about the future he said 2:59said he thinks agent-based workflows are 3:01coming next year and that it will be as 3:04big a moment for the world as the launch 3:06of the original chat GPT because think 3:08about it you give a task to the agent 3:11you go to bed the agent is done with it 3:13by morning and you have I don't know a 3:15working piece of software some large 3:17project that you want done and it's just 3:18completely finished so it's really going 3:20to change our expectations for 3:22productivity I'm sure it's going to 3:24change their pricing model I don't 3:25expect that to come cheap and it's going 3:28to change how we think about how humans 3:30interact with this kind of intelligence 3:32and the kinds of problems we solve 3:34there's going to be even more fear 3:36uncertainty and doubt in the labor 3:38market around how these agents can work 3:40as employees versus 3:42now but I don't view it as an apocalypse 3:46for jobs I actually think that these 3:48agents will be additive I think that 3:49they're going to help humans solve 3:51problems for 3:52humans and I think net net we're going 3:55to be able to work in partnership with 3:56intelligence like this to solve 3:58meaningful problems I think that's the 3:59most likely outcome it's not the only 4:01possibility like there's other 4:03possibilities out there that are that 4:04are darker for sure but looking at how 4:09ragged edged the adoption is already for 4:12chat GPT looking at how two-thirds of 4:15American workers still don't use chat 4:17GPT daily or 4:20weekly there's there's so much left room 4:23to run room to run just on adoption 4:25right and you add agents to the workflow 4:27and and there will be that same adoption 4:28curve layered on top of agent based 4:30workflows I see the pictures of the 4:32future through this Dev day but I also 4:35think it is going to arrive in fits and 4:38starts it's going to arrive in pieces 4:39for individual people and there will be 4:41lots of room to adapt if you're hungry 4:43to adapt um and as part of why I'm doing 4:46my Maven courses I think there is 4:47opportunity to adapt I don't view this 4:49as a job's Apocalypse at all one last 4:51takeaway I think this timing is really 4:53interesting with the raise that they're 4:54doing so Sam is doing a$6 a half billion 4:58dollar raise terms leaked out today that 5:01are sweeter than the original terms and 5:03I think that's influenced by the 5:05departure of technical Talent like mea 5:07Mora um and they need to be able to 5:11convince investors that they can do this 5:13and that the investors have some 5:14protection and not surprising the terms 5:16that leaked offer more protection to 5:18investors they offer an interest rate to 5:20investors they offer a clawback option 5:22Etc 5:24um and so basically investors are going 5:27to get some guarantees if Sam has trou 5:31converting this to a for-profit entity 5:33and getting the technical Talent on 5:36board to really make a run at this AGI 5:38goal that they have right artificial 5:40general 5:41intelligence we will see my suspicion is 5:44that if he can successfully raise six 5:46and a half billion he will be able to 5:47scare up the technical Talent he needs 5:49to because he'll have the salaries to 5:51command 5:53but the best people in the world are 5:55also very very scarce in this field and 5:59they many of them are already Founders 6:01and so it may look like a bunch of Aqua 6:03hires um we'll have to see I think 6:06that's one of the really interesting 6:07questions coming out of this is that the 6:08Ambitions of the company are clear it's 6:10unclear if the technical Talent they 6:12have on board with mea's departure and a 6:13few other departures is enough to get 6:16them there or if they have a talent 6:18problem that they're going to need to 6:19close uh and that's something that I'm 6:20going to be keeping my eye on we'll see 6:22maybe agent based work closer coming 6:23next year