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

    Andrew Curran Says Fable 5.1 and Mythos 5.1 Are Same Model

    Andrew Curran states the variants share identical weights but use different safeguards.

    WB
    AC
    DF
    12 Sources, 29d ago, first seen 29d ago

    TLDR

    Andrew Curran posted that Claude Fable 5.1 and Claude Mythos 5.1 are the exact same model with different safeguards. The post links to Anthropic's system card page and includes an image showing the card title. No independent confirmation of the claim appears in the packet. The message comes from an independent AI commentator whose account focuses on curating AI news and analysis.

    Combined views

    424.2K

    12 Sources, first seen 29d ago

    Combined views

    424.2K

    12 Sources, first seen 29d ago

    3.3K likes
    3.3K likes
    102 comments
    668 saves
    98 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    102 comments
    668 saves
    98 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

    12 Sources

    @AndrewCurran_System Card: Fable 5.1 and Mythos 5.1. They are the exact same model, with different safeguards. https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card
    @nsthoratThese models are really freaking good.
    @eliebakouchFable and Mythos 5.1 are the EXACT same weights, they look at the internal activations then escalate to a bigger classifier and ultimately fallback to Opus 4.8 if the request is categorized as dangerous Fable is not distilled version of a larger Mythos model
    @bilawalsidhu@alexalbert__ Congrats! So far so good. Is the cyber classifier tuned to not fire as much?
    @DanielleFongRT @doodlestein: People love to talk shit about Anthropic, but when the new frontier model drops, we all know that they’re excited as hell…
    @nrehiew_Mythos and Fable are the exact same model. So the difference in configuration is likely the threshold set for the safeguard classifier
    @MTSliveSITUATION EXPLAINED: Fable 5.1 and Mythos 5.1 are out, with the strongest cyber capabilities Anthropic has ever released. • Mythos 5.1 developed full working exploits in 98% of trials, up from 88%. Anthropic says these are the strongest cyber capabilities of any model it has released • 52.6% on Terminal Bench Science, more than double Fable 5, and evaluated with production safeguards on, so it understates the real capability • Pricing is unchanged at $10 and $50 per million tokens, but a typical workload costs 25% less and a highly agentic one costs half as much • Enterprise Frontier Safeguards store data in cloud the customer controls. Anthropic gets the alert category and severity but never the customer data • It trained a neural network to remap a third of Venus at two to three kilometer resolution, released under Creative Commons • Honesty regressed under system prompt pressure, making it the least honest Claude since Mythos preview • Anthropic's own training environments were teaching the model to hack @theojaffee: "The coding frontier might be starting to see diminishing returns in the current regime. But agentic scientific research and business workflows are starting from a much lower baseline, and so they can be improved much more rapidly." @schisofrenia: "I think laziness is kind of an inherent alignment problem, because reward hacking is a form of laziness. It's trying to find shortcuts to things. That's inherently a human and model alignment problem."
    @BrianRoemmelehttps://x.com/i/article/2094945805413298176
    @willcbgod this model is nuts. they really just made it smarter and better at coding. it can just do things. they made it reasonable and not slop. this is so cool

    12 Sources

    @AndrewCurran_System Card: Fable 5.1 and Mythos 5.1. They are the exact same model, with different safeguards. https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card
    @nsthoratThese models are really freaking good.
    @eliebakouchFable and Mythos 5.1 are the EXACT same weights, they look at the internal activations then escalate to a bigger classifier and ultimately fallback to Opus 4.8 if the request is categorized as dangerous Fable is not distilled version of a larger Mythos model
    @bilawalsidhu@alexalbert__ Congrats! So far so good. Is the cyber classifier tuned to not fire as much?
    @DanielleFongRT @doodlestein: People love to talk shit about Anthropic, but when the new frontier model drops, we all know that they’re excited as hell…
    @nrehiew_Mythos and Fable are the exact same model. So the difference in configuration is likely the threshold set for the safeguard classifier
    @MTSliveSITUATION EXPLAINED: Fable 5.1 and Mythos 5.1 are out, with the strongest cyber capabilities Anthropic has ever released. • Mythos 5.1 developed full working exploits in 98% of trials, up from 88%. Anthropic says these are the strongest cyber capabilities of any model it has released • 52.6% on Terminal Bench Science, more than double Fable 5, and evaluated with production safeguards on, so it understates the real capability • Pricing is unchanged at $10 and $50 per million tokens, but a typical workload costs 25% less and a highly agentic one costs half as much • Enterprise Frontier Safeguards store data in cloud the customer controls. Anthropic gets the alert category and severity but never the customer data • It trained a neural network to remap a third of Venus at two to three kilometer resolution, released under Creative Commons • Honesty regressed under system prompt pressure, making it the least honest Claude since Mythos preview • Anthropic's own training environments were teaching the model to hack @theojaffee: "The coding frontier might be starting to see diminishing returns in the current regime. But agentic scientific research and business workflows are starting from a much lower baseline, and so they can be improved much more rapidly." @schisofrenia: "I think laziness is kind of an inherent alignment problem, because reward hacking is a form of laziness. It's trying to find shortcuts to things. That's inherently a human and model alignment problem."
    @BrianRoemmelehttps://x.com/i/article/2094945805413298176
    @willcbgod this model is nuts. they really just made it smarter and better at coding. it can just do things. they made it reasonable and not slop. this is so cool