
context is everything

@openai | prev @bluebookhq (YC W25), 3×Founder & CTO

context is everything

Come grab me today at the lounge next to the builders stage @SlushHQ

I linked the api prompt guide earlier, and while it has a lot of great content in it, its wisdom is already baked into the system prompt in codex! If you’re looking for best practices for working with 6.2 Astra and Sol in codex, give this a read instead. learn.chatgpt.com/guides/pro…

GPT-6.2 Astra ranks #1 on AA-3D, our benchmark measuring how well models can build interactive 3D experiences. Humans preferred these scenes in 139 of 152 head-to-head votes against other models

The new Astra for Law (built on 6.2 Astra) leads on legal intelligence! Now available for all ChatGPT Enterprise customers via the Trusted Access Program. openai.com/index/astra-law-6… Absolutely GOATed work by the research team to get the pareto curve to look like this:

Thought @kimmonismus made a cool skill and it shows the jump 6.2 Astra has over 6 in visual quality even with my low effort prompt

6.2 Astra is for your most ambitious technical work, so I built Singapore. Like, the whole island 🇸🇬 100+ Astra Codex threads 1,000+ subagents Over 100 hours of work. 200,000+ building footprints 160,000+ road segments 120,000+ streetlights 1:1 scale @OpenAIDevs

what the FUCK
GPT-6.2 Astra—a frontier model built for your hardest projects. It sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT‑6 Astra. You’ll see the difference in things like slideshows, websites, and 3D designs.


GPT-6.2 Astra is our new flagship model, delivering a step up in capability with cleaner code, better frontends and 3D models, and more efficient tool use. GPT-6.2 Astra and Sol get better results on demanding tasks with fewer output tokens. In the API, Astra’s standard input and output token prices are 20% lower than GPT-6 Astra’s. Sol is more capable at the same per token API cost as GPT-6.1 Sol.



Come work with me in Stockholm 🇸🇪 We’re hiring an Applied AI Engineer to build with startups across the Nordics. You get to build alongside some of the most ambitious founders and teams in the region. Small team and room to make the role your own.

cool demo @triekki

Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world. openai.com/index/introducing…

Lots of chatter lately about memory systems and how much they cost to run. If you’re building on @OpenAI, here are a few practical tips to keep the cost down: 1. Use a 12-hour cache TTL. Handy for proactive agents. If you’re running a cron job every few hours to “dream” over past sessions, keep the shared instructions and tool definitions at the start of the prompt. A 12-hour cache write costs 2.5× the usual input token rate, but pays off over time. 2. Use Flex. Not latency sensitive task, so it’s a good place to trade speed for cost. On supported models, Flex cuts token prices by 50%. 3. Factor in the work memory saves. If an agent remembers which tool worked last time and what it needed from the result, it can skip a few calls and pull less context into the next run. Across repeated tasks, that can mean better answers, faster responses and lower overall cost, even after you count the dreaming. Rare quality + cost + latency benefit!