29 stories tagged by Digg
AI
A post announcing fidian’s TerminalBench-Fn leaderboard says GPT-6 Sol leads because models such as Astra and Fable refuse some cybersecurity tasks. It recommends GPT-6 Luna for value.
AI
A user says “we’ll raise $30m instead” if they get access to Astra ultrafast. In a reply, they say they prefer OpenAI’s startup image to Anthropic’s, though OpenAI’s models don’t work so well for their use cases.
AI
A post credits Astra with making the rewrite possible and says Dots, Space and Codex Cloud will come to web and desktop on the same day.
Technology
OpenAI says the model improves on GPT-6 Sol in coding, computer use and complex professional work, approaching GPT-6 Astra on several benchmarks. It lists cached input at $0.10 per million tokens.
AI
A user says an ‘Ultrafast mode’ is coming to GPT-6 Astra. An earlier post from the same user reads ‘GPT-6.1-Sol’ without explaining the term in its text.
AI
A user contrasts the claimed Bel release with a claim that Anthropic says it won't release models more capable than Fable.
AI
A user suggests that Claude gives Astra a concrete reference to a strong non-GPT AI, which might help it get outside its “basin” without seeing any of Claude’s work.
AI
A user says Sonnet 5.5 is better than Astra and has canceled their OpenAI subscriptions. They praise Opus 5.5 but argue that Sonnet’s high output-token use eats into its speed and cost gains.
AI
One user says Opus 5.5 is working well for agentic coding tasks they used to handle with Fable 5.1, while noting they hadn’t tested it as rigorously.
AI
A user says Astra beat Zork like someone with a walkthrough, rather than someone experiencing the game for the first time.
AI
A user argues that harder document processing benefits from high parameter space. They say they rely mostly on Astra for initial annotation and use Gemma 4 31b for fine-tuning.
AI
The Wall Street Journal says the model was slated for ChatGPT and Codex. CNBC says OpenAI found it fell short of the company’s safety standards.
AI
A user says Sonnet 5.5 performs better than GPT 6 sol on many benchmarks and is on par with Astra.
AI
The user who ran the test says Opus 5.5’s physics and visual quality wasn’t noticeably worse. They suggest line count could be a rough measure of code elegance, though not a perfect one.
AI
A user says Opus 5.5 on a $100 Claude plan lasts them six to seven days a week, compared with one to 1.5 days for Astra on a $200 Codex plan.
AI
One poster says OpenAI seems unworried by Opus 5.5 and speculates that an Astra update, or at least a reveal of its successor, could come soon.
AI
A user says they asked Astra to turn their photo collection into an app for browsing it in different ways. They also say Luna reviewed and labeled the photos for filtering and categorization.
World & Business
A business owner says free AI tools help them work without a huge team, pairing Astra with Vercel for website deployment and using GitHub to track and roll back changes.
Science
Posts quote Sam Altman claiming a leap in math capabilities beyond Astra and predicting that AI will transform biology and materials science.
Technology
OpenAI says the framework sets criteria and timelines for public disclosure, including cases it hasn't fully explained or mitigated. More complex cases may require longer investigations.
AI
OpenAI co-founder highlights Astra tool for everyday personal and work tasks.
Technology
The approach improves performance but complicates inspection by researchers.
AI
Noam Brown says OpenAI announces internal model math results only when they shift views on AI progress.
AI
OpenAI slows frontier training runs after models show misalignment and to add safeguards.
AI
OpenAI shares preliminary evaluations for Astra and steps to strengthen safeguards against critical cyber capabilities.
AI
X users share reports from a leaker about an upcoming OpenAI model release.
AI
Ethan Mollick shares a math professor's take on AI progress.
AI
Cognitive scientist compares AI demo to historical math advances like calculus and algebra.
AI
Internal Astra model produces Lean-certified proofs for ten open problems across math and theoretical computer science.
