Anthropic has introduced the “most agentic model” of all existing in this line. What can Claude Sonnet 5
Anthropic has introduced the new Claude Sonnet 5, which it positions as the most agent-oriented model of all existing models in this line.
Anthropic has introduced the new Claude Sonnet 5, which it positions as the most agent-oriented model of all existing models in this line.
As the Anthropic blog notes, the model can «make plans,» use browsers and terminals on its own, and «operate autonomously at a level that only a few months ago required larger and more expensive models.»
The company notes that the Claude Sonnet 5 is approaching Opus class models in terms of agent capabilities.
«Sonnet 5 closes the gap: its performance is close to Opus 4.8, but at a lower price. This is a significant improvement over its predecessor, Sonnet 4.6, in important aspects of agent performance,» Anthropic claims.
There they provided a table of the Claude Sonnet 5's performance compared to other models.
It is noted that Claude Sonnet 5 is now available for all plans. It is the default model for Free and Pro plans, and is also available for Max, Team, and Enterprise users.
«It is also available on Claude Code and the Claude platform, where its initial price is $2 per million input tokens and $10 per million output tokens until August 31, 2026, after which its price will be $3 per million input tokens and $15 per million output tokens.
Developers can use Claude Sonnet 5 through the Claude API.
The AI giant adds that the previous top-of-the-line Sonnet model (Sonnet 4.6) lagged significantly behind the Opus 4.8, but now the Sonnet 5 and Opus 4.8 cover the same range, with the Sonnet 5 offering «impressive capabilities at a lower price, while the Opus 4.8 offers greater accuracy at a higher price.»
In terms of agent security, the model, according to the company, is better at rejecting malicious requests and resisting «capture attempts.» The model also exhibits lower rates of hallucinations and «smearing» than the Sonnet 4.6.
Sonnet 5 exhibits a generally lower level of misbehavior than Sonnet 4.6, although higher than Mythos Preview and Opus 4.8.
Regarding the use of this model for cybersecurity tasks, the company said that it intentionally did not train Sonnet 5 on these tasks, although it «can perform some routine, harmless cyber tasks.» However, in developing software exploits, the model shows significantly worse performance than Opus 4.8 and Mythos 5.
As dev.ua reported, OpenAI has introduced three new models, including the flagship Sol. However, in accordance with the new requirements of the Trump administration, they are currently in «limited preview.»
Base44 vibecoding platform launches its own AI model
Base44 vibecoding platform launches its own AI model
Claude Mythos 5 is now available to over 100 companies and institutions. The US government has relaxed restrictions on this model Anthropic
Claude Mythos 5 is now available to over 100 companies and institutions. The US government has relaxed restrictions on this model Anthropic
Read the country’s main IT news in our Telegram
Read the country’s main IT news in our Telegram
