GPT-6 Astra: an automated AI Engineer you can hire for <$6 an hour

OpenAI launched GPT-6 Astra, a supermodel that beats Fable 5.1 on benchmarks including FrontierMath (97.6%) and ARC-AGI-3 (99.9%), and is capable of acting as an automated AI engineer for under $6 an hour, according to early testing by Latent Space. The model can choose and train models, label data, deploy systems, and manage subagents over billions of tokens, with OpenAI's Greg asserting AGI is here.

GPT-6 Astra , the first Stargate https://x.com/ZeffMax/status/2095582617035063375?s=20 and lightly looped https://x.com/rasbt/status/2095141254958858496 supermodel from OpenAI, launched today http://GPT-6 Astra’s launch , cleanly beating Fable 5.1 https://www.latent.space/p/ainews-claude-fablemythos-51-new on many metrics including completely saturating the hardest versions of FrontierMath https://buttondown.com/ainews/archive/ainews-frontiermath-a-benchmark-for-evaluating/ 97.6% and ARC-AGI-3 https://openai.com/index/gpt-6-astra/ citation-top-1 99.9% . Lots of demos will focus on typical talk tracks like the computer use https://x.com/OpenAI/status/2095595741528125780 to the Pokemon playing https://x.com/Clad3815/status/2095596013168050551 to Blender https://x.com/tomkrcha/status/2095598645190291775 to the scientific https://x.com/polynoamial/status/2095583211950833768 and cybersafety https://openai.com/index/gpt-6-astra/ citation-top-13 benchmarks system card https://deploymentsafety.openai.com/gpt-6-astra . Greg says AGI is here https://x.com/ZeffMax/status/2095582614648447179 , and Jakub says it is finally the Automated AI Research Intern https://www.latent.space/p/ainews-openai-to-reach-agi-bar-by he wanted. We aren’t qualified to talk about those, but we got early access and threw it at every practical, real-life task we could think of. After burning over 20B tokens of Astra , we can confirm the most surprising finding: GPT-6 Astra is one of a new class of models 1 footnote-1 that are fully capable AI Engineers in their own right . They now help you choose and train models , label data both helping you label and then using your labels for active learning, like SAM https://www.youtube.com/watch?v=sVo7SC62voA , keep pipelines saturated , instrument and read logs , deploy and debug entire systems in one shot, fan out and command and eval subagents including agents running other models , and keep coherence over billions of tokens of a single agent thread. Raising Your Ambitions We’ve written before about the high-return activity of raising your aspirations for LLMs https://www.latent.space/p/ainews-the-high-return-activity-of?utm source=publication-search . Our experience has made us exponentially more ambitious than we have ever been . Over the past month, we went from prompting humans for a fun “ Kill My SaaS https://x.com/swyx/status/2085517544795079014 ” competition 2 footnote-2 , to building a dozen internal/personal tools https://tools.aieconf.com/ , including 4 previously paid SaaS tools https://swyx.io/tools , fully redesigned my personal site https://swyx.io/ , made an incomplete but functional replacement of GitHub + Vercel https://forge.smol.ai/ , trained game AI for a strategy board game with 10,000x more legal moves than Go https://overgrid.swyx.io/ ai-rivals , saved tens of thousands of dollars in personal finance cleanups, republished my old book https://learninpublic.org/ with synced audiobook audio and printed physical editions, and even more https://aeo.latent.space/ ambitious https://news.latent.space/ projects we will launch soon. The $6 an hour number might sound surprising, but that’s exactly what we saw in our testing https://aeo.latent.space/ operations - 33 tokens per second at a max $50 per million token rate. Given that Astra is more token efficient than Sol and Fable independently confirmed by Artificial Analysis https://x.com/ArtificialAnlys/status/2095595494081024077 , it often means that Astra is simultaneously also the best fast-and-smart model you can buy assuming our preview latency holds for GA , outside of Spark 1.3 https://www.latent.space/p/ainews-muse-spark-13-matches-gpt . Managing fleets of subagents individually tweaked, bounded concurrency Now of course, if you just throw on Astra at Ultra you’re gonna burn through a lot more than $6 per hour…. because it is so dang good at parallelizing. Monitoring its own runs, starting and stopping waves This is basically what i’d pay a junior AI Engineer to do. and I spent about $100 over 2 days to do this. Making model benchmarks, handling money, making estimates, scaling up runs, getting human ratings Wild. We are out of time for this writeup, will complete this later, so if you are reading this on email, check back at the end of the day. However the overall conclusion you should have is that OpenAI have clearly trained a model that is capable of automating much of their own AI Engineering, and it is finally time that you learn to exploit Astra- and Fable-class models and be far, far more unreasonable https://www.youtube.com/watch?v=9fubhllmsBU&pp=0gcJCUAdAYcqIYzv . 1 footnote-anchor-1 We are running similar work https://aeo.latent.space/ on Grok, Fable and other similar frontier models but OpenAI was most generous with trial limits so this gets the writeup - but the agentic coding patterns discussed here will likely apply to all such late 2026 frontier models. 2 footnote-anchor-2 Many of you are waiting to hear results… sorry for the radio silence we got… busy We will announce winners and reimbursements and best attempts.