
Dominus
5.1K posts

Dominus
@Dominus_Crypto
TG channel: https://t.co/gK06YhvT2D




🚨 CRAFTY CAPYBARA ON THE RUN FOR NINE DAYS – ZOOKEEPERS LAUNCH TASTY MELON TRAP TO CATCH THE GIANT RODENT! Samba legged it from Marwell Zoo in Hampshire just ONE DAY after arriving on March 16, squeezing under a gate with mate Tango (who was quickly recaptured in nearby bushes). The adventurous female has been spotted sunbathing on the River Itchen riverbank in Twyford, hiding in waterways and staying active at night while keepers hunt her with thermal drones and camera traps. Now they’ve set humane badger traps baited with her favourite MELON along the river in a last-ditch “tasty plan” to lure her home. Still at large on day 10. Britain’s most wanted capybara! 😂😂

LEAK: Anthropic’s new model “Claude Mythos” (Capybara tier) just surfaced via ~3,000 exposed internal assets found in a public cache. It’s a brand new class, larger than Opus, described as a “step change” and their most powerful model yet. “Compared to Claude Opus 4.6, Capybara achieves dramatically higher scores in software coding, academic reasoning, and cybersecurity.” Anthropic says it’s already “far ahead” in cyber capabilities, so strong they fear misuse and may give defenders early access first. Per Dario’s earlier hints, this could be a ~10T parameter model that cost ~$10B to train.


@nickcammarata 😂♥️


LEAK: Anthropic’s new model “Claude Mythos” (Capybara tier) just surfaced via ~3,000 exposed internal assets found in a public cache. It’s a brand new class, larger than Opus, described as a “step change” and their most powerful model yet. “Compared to Claude Opus 4.6, Capybara achieves dramatically higher scores in software coding, academic reasoning, and cybersecurity.” Anthropic says it’s already “far ahead” in cyber capabilities, so strong they fear misuse and may give defenders early access first. Per Dario’s earlier hints, this could be a ~10T parameter model that cost ~$10B to train.



Ejaaz @cryptopunk7213 caught this. Anthropic accidentally leaked their next model. They're calling it "by far the most powerful AI model we've ever developed." Codename: Capybara. Internal name: Mythos. Released date stamped March 2026. It wasn't supposed to be public. A researcher's dev environment briefly exposed the research preview page. The internet noticed immediately. What the leak says. The leaked page shows two versions side by side — v1 Mythos and v2 Capybara — under the label "Research preview." The language is unambiguous: "We have finished training a new AI model: Claude Mythos. It's by far the most powerful AI model we've ever developed." And the performance claim: "Compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others." Dramatically higher. Not marginally. Not incrementally. Dramatically.

Anthropic 自己发了篇博客,又自己撤了。 但截图已经传遍全网了。 Claude Mythos,比 Opus 4.6 高一个层级,官方新设了个 Capybara 档。编码、推理、网络安全全面碾压。 最有意思的是 Anthropic 自己的态度——他们说这个模型的网络安全能力太强了,先只给防御机构用,让大伙赶紧把代码加固了,不然以后 AI 驱动的攻击谁都挡不住。 翻译成人话就是:我造了把刀,很锋利,我先给保镖用,你们赶紧换门。 安全公司股价闻讯暴跌 4%。 所以 AI 发展到现在,已经不是"能不能写代码"的问题了,是"能不能黑掉你的代码"的问题。 Opus 刚觉得够用了,Mythos 就来了。Opus 还没降智呢,Mythos 就已经让你降智了。 #ClaudeMythos #Anthropic #AI










