• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Analyst Links Astra Training to May-July Usage Spikes

    Posts cite internal usage spikes as evidence for GPT-6 Astra training timeline.

    T(
    EL
    LA
    9 Sources, 24d ago, first seen 24d ago

    TLDR

    @scaling01 posted a chart titled Experiment velocity has increased and stated that training of Astra likely occurred between May and late July. The account further claimed GPT-6 Astra was likely broadly available to OpenAI employees between July 12 and July 26, citing an 87 percent rise in median spending, a 169 percent rise in 90th-percentile spending, a 33 percent rise in researcher output-token usage, and a 33 percent rise in agent runtime during that window.

    Combined views

    137.7K

    9 Sources, first seen 24d ago

    Combined views

    137.7K

    9 Sources, first seen 24d ago

    1.1K likes
    1.1K likes
    52 comments
    197 saves
    39 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    52 comments
    197 saves
    39 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    9 Sources

    @scaling01this is further confirmation for me that training of Astra likely happened between May and late July
    @eliebakouchhere is what it looks like if we correct by astra pricing. note that token consumption alone wouldn't be a good metric either since you pay more per token but use fewer tokens and get more intelligence, still an interesting and different view though > For cost, we use retail API prices at the processing tier most similar to the internal deployment. All prices are as of September 3, 2026. would be very weird to include promotions imo, but astra told me that the language used by oai means they included them so i put that on the plot as well also the median daily $ / researcher in february is close to 0 at openai which is super surprising?? sota oai public model in Feb 2026 was GPT 5.3 codex
    @zephyr_z9Well, using API costs as a metric is wrong, especially when they have 90%-95% inference margins The actual spend (reported in their OPEX) is currently $22k - $250k+ per year per researcher
    @teortaxesTexView post on X

    9 Sources

    @scaling01this is further confirmation for me that training of Astra likely happened between May and late July
    @eliebakouchhere is what it looks like if we correct by astra pricing. note that token consumption alone wouldn't be a good metric either since you pay more per token but use fewer tokens and get more intelligence, still an interesting and different view though > For cost, we use retail API prices at the processing tier most similar to the internal deployment. All prices are as of September 3, 2026. would be very weird to include promotions imo, but astra told me that the language used by oai means they included them so i put that on the plot as well also the median daily $ / researcher in february is close to 0 at openai which is super surprising?? sota oai public model in Feb 2026 was GPT 5.3 codex
    @zephyr_z9Well, using API costs as a metric is wrong, especially when they have 90%-95% inference margins The actual spend (reported in their OPEX) is currently $22k - $250k+ per year per researcher
    @teortaxesTexView post on X