Breaking
KRAn Se-young Aims for Back-to-Back Asian Games Badminton Gold as South Korea Competes in Multiple FinalsFRFrance wins in Belgium thanks to Michael Olise and takes the lead in Group A of the League of NationsDERising oil prices and high bond yields are weighing on Wall StreetBRPolice save newborn who choked on milk in Mato Grosso do SulBRNGO Lar Espírita Maria Lobato de Freitas is invaded and robbed in UberlândiaCRYPTO-ENCrypto king Aiden Pleterski to stand trial in Canada for $30 million fraud schemeUSEagles Seek Fourth Straight Road Win Against Bears in Monday Night Football MatchupBRTV Globo promotes debate between candidates for the government of Rio de Janeiro after soap operaRUIran is not negotiating its nuclear programJPChina's Ministry of Foreign Affairs asks former Foreign Minister Iwaya to correct his incorrect statements on the Taiwan issue, and multiple news reports include the election of the first female president of the University of Tokyo, and progress in AI and trade talks between Japan, the U.S., and China.KRAn Se-young Aims for Back-to-Back Asian Games Badminton Gold as South Korea Competes in Multiple FinalsFRFrance wins in Belgium thanks to Michael Olise and takes the lead in Group A of the League of NationsDERising oil prices and high bond yields are weighing on Wall StreetBRPolice save newborn who choked on milk in Mato Grosso do SulBRNGO Lar Espírita Maria Lobato de Freitas is invaded and robbed in UberlândiaCRYPTO-ENCrypto king Aiden Pleterski to stand trial in Canada for $30 million fraud schemeUSEagles Seek Fourth Straight Road Win Against Bears in Monday Night Football MatchupBRTV Globo promotes debate between candidates for the government of Rio de Janeiro after soap operaRUIran is not negotiating its nuclear programJPChina's Ministry of Foreign Affairs asks former Foreign Minister Iwaya to correct his incorrect statements on the Taiwan issue, and multiple news reports include the election of the first female president of the University of Tokyo, and progress in AI and trade talks between Japan, the U.S., and China.
BackAnthropic launches Sonnet 5.5, a faster and lower-cost AI model focused on safety and alignment
Anthropic launches Sonnet 5.5, a faster and lower-cost AI model focused on safety and alignment
Developing
CNBC3 hours agoTech2 min read

Anthropic launches Sonnet 5.5, a faster and lower-cost AI model focused on safety and alignment

Quick Look

Anthropic released Sonnet 5.5, a faster and lower-cost AI model better at coding and document creation than its predecessor, launched less than a week after Opus 5.5 and ahead of Haiku 5.5, with pricing at $2 per million input tokens and $10 per million output tokens, emphasizing safety alignment despite not advancing model capabilities frontier.

AI-generated summary

Why It Matters

Anthropic has been under scrutiny for AI safety practices following researcher warnings about catastrophic harm risks, prompting CEO Dario Amodei to urge AI companies to slow development of advanced models.

Font size

Anthropic on Monday released a new artificial intelligence model called Sonnet 5.5, the company's second launch since CEO Dario Amodei called to slow down the pace of development.

Sonnet 5.5 is a faster, lower-cost offering from Anthropic, and the company said it is better at coding, completing scoped tasks and creating polished documents, slides and spreadsheets than its predecessor. The launch comes less than one week after Anthropic introduced Opus 5.5, a more expensive model, and it said Monday that Haiku 5.5, the cheapest offering in the suite, is coming soon.

"Sonnet is really for the cost-conscious customer where they might not need as much intelligence," Theo Chu, a research product manager at Anthropic, told CNBC in an interview. "It might be routine tasks that just need execution, but don't need that judgment that Opus can bring."

Anthropic's safety and security practices, as well as those of its competitors, have been under intense scrutiny in recent weeks after a growing chorus of researchers warned about AI's potential to cause catastrophic harm. In response, Amodei stunned the tech sector and government officials earlier this month by urging AI companies to slow how quickly they improve their most advanced models.

The company said Monday that Sonnet 5.5 doesn't advance the frontier of its models' capabilities, so most of its alignment testing focused on a "targeted set of risks that apply to models of any capability level," according to a blog post. In the AI industry, an aligned model is one that behaves in accordance with human interests and values.

However, Anthropic said that Sonnet 5.5 has cybersecurity capabilities that are a "large improvement" over those of its predecessor, so it's the first Sonnet model to launch with similar fallbacks and cyber safeguards that Anthropic has developed for its most capable models.

"Focusing on alignment and safety has been a key part of our mission from the very beginning," Chu said. "This is something that we've always prioritized across our models."

Sonnet 5.5 is available at $2 per million input tokens, and $10 per million output tokens, half the price of Opus 5.5. Anthropic said Sonnet 5.5 also requires fewer tokens per task than its previous version, which makes it cheaper to run.

What to Watch

AI outlook — possibilities, not facts

  • Haiku 5.5 will be released soon as the cheapest offering in Anthropic's model suite

    Very likely · Within weeks

Open Questions

  • When will Haiku 5.5 be released?
  • How does Sonnet 5.5's performance compare to competing models from other AI companies?
  • What specific cybersecurity improvements does Sonnet 5.5 offer over its predecessor?

Related Topics

This article was originally published by CNBC.

Related Stories

The first manned flight of Starliner to the ISS has been postponed to mid-2028
Developing·

The first manned flight of Starliner to the ISS has been postponed to mid-2028

NASA and Boeing announced the postponement of the first scheduled operational flight of the Starliner spacecraft with a crew to the ISS to mid-2028. An uncrewed Starliner 1 mission is scheduled no earlier than December 2026 to confirm the results of the troubleshooting. About $359 million has been allocated for the modification of engines, on-board systems and certification of the Vulcan Centaur launch vehicle.

РИА Новости
1 min read
Dodo Pizza suffered a cyber attack and customer data may have been compromised
Developing·

Dodo Pizza suffered a cyber attack and customer data may have been compromised

The pizzeria chain Dodo Pizza reported a cyber attack on its IT system, as a result of which the names, addresses, emails, telephone numbers, dates of birth and order contents of some customers could be accessed. Payment information remains secure as it is not stored in the system. The company notified Roskomnadzor and blocked access to the attackers; an internal audit is underway.

РИА Новости
1 min read
More on this topicanthropic