1. X
  2. Dan McAteer
Log inSign up
Dan McAteer
34.9K posts
user avatar
Dan McAteer
@daniel_mac8
I build AI agents for a living and write about what they're doing to our minds. Essays weekly at attentionheads.blog · Agentic Ops @ AnswerRocket
United States
attentionheads.blog
Joined October 2013
4,482
Following
27.1K
Followers
RepliesRepliesArticlesArticlesMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • Pinned
    user avatar
    Dan McAteer
    @daniel_mac8
    6h
    Codex users: here's an amazing way to take advantage of the efficiency and capability of GPT-5.6 Luna Max. It's called 'sol-advisor'. 1. GPT-5.6 Sol High as orchestrator 2. GPT-5.6 Luna Max as implementer for routine tasks 3. GPT-5.6 Terra Max as implementer for complex tasks
    00:00
    00:19
    user avatar
    Dan McAteer
    @daniel_mac8
    Jul 31
    Codex users: do this right *now*. Max reasoning effort is off by default. Luna at Max reasoning is ~ Sol Medium / Opus 5 Medium level at 1/6th the cost. Change your life today.
    138K
  • user avatar
    Dan McAteer
    @daniel_mac8
    34m
    Wow. Lotta love for ‘sol-advisor’. Uses GPT-5.6 Sol as orchestrator/advisor and Luna and Terra on Max reasoning as implementers, with their 80% & 20% price reductions. Luna Max is Sol Medium level and Terra Max is Sol High level on AA-Agentic coding. Graph from @rasbt
    00:21
    user avatar
    Dan McAteer
    @daniel_mac8
    6h
    Codex users: here's an amazing way to take advantage of the efficiency and capability of GPT-5.6 Luna Max. It's called 'sol-advisor'. 1. GPT-5.6 Sol High as orchestrator 2. GPT-5.6 Luna Max as implementer for routine tasks 3. GPT-5.6 Terra Max as implementer for complex tasks
    1.1K
  • user avatar
    Dan McAteer
    @daniel_mac8
    10h
    Astra achieved 10 scientific breakthroughs for a total of $2,000. These are problems that human scientists were unable to solve, at any cost. At the same time, OpenAI has found major test time efficiency improvements, dropping the cost of Luna and Terra 80% and 20%,
    user avatar
    Noam Brown
    @polynoamial
    14h
    Replying to @polynoamial
    And yes we did try other major problems without success. Sadly no Millennium Prize problems (yet). But also, we didn’t spend a lot on each problem. It’s possible to push test-time compute much further.
    5.5K
  • user avatar
    Dan McAteer
    @daniel_mac8
    12h
    👀 OpenAI's next-gen AI model: Astra > Solves 10 major problems across math, quantum complexity and theoretical computer science > Costs $2,000 at GPT-5.6 Sol API prices to solve all 10 problems > Noam Brown calls it a "major step for scientific reasoning" A report yesterday
    4K
  • user avatar
    Dan McAteer
    @daniel_mac8
    23h
    Now we know what @polynoamial and team have been working on: Astra. A multi-agent system that is trained to perform well specifically on ultra-long horizon tasks. This report from The Information says OpenAI plans to announce they’ve solved 10 unsolved maths problems,
    user avatar
    Lentils
    @Lentils80
    Jul 31
    🚨 OpenAI preparing to release a new model family, tentatively named "Astra" OAI touted its abilities to have multiple agents work together over a long period of time to solve particularly hard problems Astra is a new class of OpenAI models alongside Sol, Terra and Luna h/t
    14K