Similar situation here - I went through this last year when everyone was hyping up the jump to OpenAI GPT-4 and I was trying to map out a massive security automation pipeline. Honestly, in my experience, waiting for the "next big thing" like GPT-6 can be a total trap for your budget. Over the years, I've learned that these rumors about 'Strawberry' or massive training clusters are cool, but relying on them for a 6-12 month roadmap is highkey risky because safety testing usually delays everything by months.

I remember dumping a ton of time into prepping for a release that didn't hit for ages, and it basically nuked our Q3 goals. What worked for me was focusing on reliability and safety with what's actually available and cheap right now. Like, instead of waiting for a GPT-6 that might cost a fortune per token, I started experimenting with Meta Llama 3 70B on local clusters. It's basically free if you have the hardware, or you can run it on Groq Cloud LPU Inference for pennies. It's way more budget-friendly than the $20/mo per user for OpenAI ChatGPT Plus or the high API costs of OpenAI o1-preview. Plus, keeping things local or on cheaper open-weight models is a huge win for data safety. I mean, do we even know if GPT-6 will be affordable for high-volume automation?? I'd definately suggest building your logic around current specs and just making it modular so you can swap it later if a 'miracle' model actually drops in early 2025. Just my two cents from being burned by the hype cycle before lol.

Similar situation here - I went through this last year when everyone was hyping up the jump to OpenAI GPT-4 and I was trying to map out a massive security automation pipeline. Honestly, in my experience, waiting for the "next big thing" like GPT-6 can be a total trap for your budget. Over the years, I've learned that these rumors about 'Strawberry' or massive training clusters are cool, but relying on them for a 6-12 month roadmap is highkey risky because safety testing usually delays everything by months.

I remember dumping a ton of time into prepping for a release that didn't hit for ages, and it basically nuked our Q3 goals. What worked for me was focusing on reliability and safety with what's actually available and cheap right now. Like, instead of waiting for a GPT-6 that might cost a fortune per token, I started experimenting with Meta Llama 3 70B on local clusters. It's basically free if you have the hardware, or you can run it on Groq Cloud LPU Inference for pennies. It's way more budget-friendly than the $20/mo per user for OpenAI ChatGPT Plus or the high API costs of OpenAI o1-preview. Plus, keeping things local or on cheaper open-weight models is a huge win for data safety. I mean, do we even know if GPT-6 will be affordable for high-volume automation?? I'd definately suggest building your logic around current specs and just making it modular so you can swap it later if a 'miracle' model actually drops in early 2025. Just my two cents from being burned by the hype cycle before lol.

Similar situation here - I went through this last year when everyone was hyping up the jump to OpenAI GPT-4 and I was trying to map out a massive security automation pipeline. Honestly, in my experience, waiting for the "next big thing" like GPT-6 can be a total trap for your budget. Over the years, I've learned that these rumors about 'Strawberry' or massive training clusters are cool, but relying on them for a 6-12 month roadmap is highkey risky because safety testing usually delays everything by months.

I remember dumping a ton of time into prepping for a release that didn't hit for ages, and it basically nuked our Q3 goals. What worked for me was focusing on reliability and safety with what's actually available and cheap right now. Like, instead of waiting for a GPT-6 that might cost a fortune per token, I started experimenting with Meta Llama 3 70B on local clusters. It's basically free if you have the hardware, or you can run it on Groq Cloud LPU Inference for pennies. It's way more budget-friendly than the $20/mo per user for OpenAI ChatGPT Plus or the high API costs of OpenAI o1-preview. Plus, keeping things local or on cheaper open-weight models is a huge win for data safety. I mean, do we even know if GPT-6 will be affordable for high-volume automation?? I'd definately suggest building your logic around current specs and just making it modular so you can swap it later if a 'miracle' model actually drops in early 2025. Just my two cents from being burned by the hype cycle before lol.

> knowing if a massive jump in reasoning or context window is coming in the next 6-12 months would really change our strategy

Quick question—are you guys mostly looking for better logic or just a bigger memory for your automation? Honestly, I'm kinda newer to the market research side of things, but I feel like sticking with a big brand like Google or Anthropic is a safe bet right now regardless of the exact version. But yeah, before I dive into the rumors, what specifically is your project lacking today? lol

Subbing for updates

Similar situation here - I went through this last year when everyone was hyping up the jump to OpenAI GPT-4 and I was trying to map out a massive security automation pipeline. Honestly, in my experience, waiting for the "next big thing" like GPT-6 can be a total trap for your budget. Over the years, I've learned that these rumors about 'Strawberry' or massive training clusters are cool, but relying on them for a 6-12 month roadmap is highkey risky because safety testing usually delays everything by months.

I remember dumping a ton of time into prepping for a release that didn't hit for ages, and it basically nuked our Q3 goals. What worked for me was focusing on reliability and safety with what's actually available and cheap right now. Like, instead of waiting for a GPT-6 that might cost a fortune per token, I started experimenting with Meta Llama 3 70B on local clusters. It's basically free if you have the hardware, or you can run it on Groq Cloud LPU Inference for pennies. It's way more budget-friendly than the $20/mo per user for OpenAI ChatGPT Plus or the high API costs of OpenAI o1-preview. Plus, keeping things local or on cheaper open-weight models is a huge win for data safety. I mean, do we even know if GPT-6 will be affordable for high-volume automation?? I'd definately suggest building your logic around current specs and just making it modular so you can swap it later if a 'miracle' model actually drops in early 2025. Just my two cents from being burned by the hype cycle before lol.

Similar situation here - I went through this last year when everyone was hyping up the jump to OpenAI GPT-4 and I was trying to map out a massive security automation pipeline. Honestly, in my experience, waiting for the "next big thing" like GPT-6 can be a total trap for your budget. Over the years, I've learned that these rumors about 'Strawberry' or massive training clusters are cool, but relying on them for a 6-12 month roadmap is highkey risky because safety testing usually delays everything by months.

I remember dumping a ton of time into prepping for a release that didn't hit for ages, and it basically nuked our Q3 goals. What worked for me was focusing on reliability and safety with what's actually available and cheap right now. Like, instead of waiting for a GPT-6 that might cost a fortune per token, I started experimenting with Meta Llama 3 70B on local clusters. It's basically free if you have the hardware, or you can run it on Groq Cloud LPU Inference for pennies. It's way more budget-friendly than the $20/mo per user for OpenAI ChatGPT Plus or the high API costs of OpenAI o1-preview. Plus, keeping things local or on cheaper open-weight models is a huge win for data safety. I mean, do we even know if GPT-6 will be affordable for high-volume automation?? I'd definately suggest building your logic around current specs and just making it modular so you can swap it later if a 'miracle' model actually drops in early 2025. Just my two cents from being burned by the hype cycle before lol.

Ok so, I've been tracking these dev cycles for a while and honestly... i think we need to temper the hype a bit. From a professional standpoint, jumping straight into a massive automation project based on [[PRODUCT:OpenAI ChatGPT]] leaks is super risky.

In my experience, here's what the landscape actually looks like:

- **The Naming Game**: What people call GPT-6 right now is mostly just speculation. We're likely gonna see [[PRODUCT:OpenAI o1-preview]] and more iterative updates (like the 'Strawberry' reasoning models) before a full version jump.
- **Timeline Reality**: Realistically? I wouldn't expect a public 'GPT-6' until late 2025 or even 2026. Training these massive clusters takes forever, plus the safety red-teaming is getting way more intense.
- **Stability First**: For your team, I'd suggest building on [[PRODUCT:OpenAI GPT-4o]] for now. It's solid and has the reliability you need for production.

Basically, don't wait for a ghost. Stick with what's verifiable cuz 'soon' in AI-speak usually means 'when it's ready' lol. gl with the project!

Regarding what #10 said about "Would love to know this too" - honestly, it is so frustrating that we're all just sitting here guessing. I've been messing with these models since the early days and the way these companies handle communication now is just... ugh. Its like a constant carrot on a stick. You try to plan a real project and they drop some vague teaser that tells us basically nothing about actual deployment dates. It drives me crazy how they leave developers in the dark while they play these hype games. It feels like such a scam sometimes, keeping everyone paying while they drip-feed minor updates instead of giving us a real roadmap. I had such high hopes for better transparency this year but its just getting worse. Honestly, i've had issues with the current versions just suddenly getting dumber or losing context, and it's not as good as expected for the price they charge. Quality feels like it fluctuates every week and we're just expected to roll with it. If you guys ever need to vent more about it or just want to bounce some ideas off someone who has been through this cycle a dozen times, im here. We're all in this mess together.

Sooo I was just reading through your post and honestly, I love the energy! Planning out long-term automation projects is such a smart move, but also kinda stressful when the tech moves this fast. I've been deep in the weeds with LLM infrastructure and training cluster costs lately, and it's wild how much *everyone* is speculating right now. I've been there, trying to figure out if I should wait for the next big release or just pull the trigger on what's available today... it's a total gamble!!

Before I dive into the technical roadmap stuff and what I'm seeing in the dev circles, I'm curious about one thing: what's the MAIN bottleneck for your team's current projects? Like, are you guys mostly hurting for a bigger context window to process massive docs, or is it more about needing that specialized reasoning for complex coding or logic? Knowing exactly what you're waiting for would *really* help me give you a better technical breakdown. Cheers!

Nice, didn't know that

Would love to know this too