跳到正文
Silicon Republic· Tim Barnwell·· 12 天前AI 評分70

Anthropic 在 IPO 前推出 Claude Fable 5.2 和 Claude Mythos 5.2

Anthropic launches Fable 5.2 and Mythos 5.2 ahead of IPO

AI 導讀

Anthropic 推出 Claude Fable 5.2 和 Claude Mythos 5.2,兩者是同一基礎模型的變體,並在編碼、推理和視覺方面有所改進。

正文

Fable and Mythos are designed for ‘the most ambitious, open-ended technical work’, according to their maker.

AI giant Anthropic has unveiled two new model updates that it said “are designed to tackle the most daunting technical challenges”, ahead of its planned initial public offering (IPO) later this month.

Claude Fable 5.2 and Claude Mythos 5.2, which are variants of the same base model, both show improvements in areas like coding, reasoning and vision, according to their maker.

Fable is available to all users, while Mythos is available to organisations pursuing beneficial applications in cybersecurity or biomedical research through special access programmes, Anthropic said in a blogpost.

It added that the pair of new models are both faster and cheaper than their 5.1 predecessors, which were launched in September.

Fable 5.2 and Mythos 5.2 also come with an optional ‘fast mode’, which Anthropic said speeds up token generation by up to 2.5 times and is available for $16 per million input tokens and $80 per million output tokens.

Fable and Mythos are designed for “the most ambitious, open-ended technical work”, according to their maker, while Fable 5.2 also has the ability to advise or delegate tasks to less powerful Anthropic models.

“To help ensure safety and alignment keep step with growing model capabilities, we ran extensive prerelease safety testing on the model that underlies Fable 5.2 and Mythos 5.2,” Anthropic said.

“This included developing novel evaluations that give deeper insight into specific concerning behaviours like sandbox escapes.

“We believe high-stakes risks from misalignment remain low, and the model performed as well as, or better than, our recent models on most alignment measures.”

Over the summer, Claude models including Mythos variants were involved in cybersecurity controversies after gaining unauthorised internet access during testing.

In late September, Anthropic’s rival OpenAI delayed the launch of its planned newest model GPT-6.1 Astra due to security concerns.

The company also committed to setting up a taskforce in Australia to “develop practical policy recommendations” for managing risks from AI agents as a response to a June incident in which its models hacked into Australia’s Medicare website.

Anthropic, meanwhile, has tapped Accenture to independently test its frontier AI developments in a first-of-its-kind third-party security evaluation.

Anthropic’s imminent IPO has the potential the value the company at more than $1.5trn, it has long been speculated. OpenAI has delayed its planned IPO into 2027 over safety concerns and public sentiment, according to its CEO Sam Altman.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

來源:Silicon Republic · siliconrepublic.com