This is an info Alert.
⌘K
  • Home
  • News
  • Blog
  • Releases
  • LLM history
  • Compare LLMs
  • Library
  • About
Sign in

A blog and notes on development. The easiest way to reach me is via the social links below.

Documents
Terms of UsePrivacy Policy
Contacts
talalaev.misha@gmail.com

© All rights reserved.

Anthropic in 5 Days: Sonnet 5 as the Default, the Return of Fable 5, and New Security Measures

Mikhail T. (Sh0ny)
Mikhail T. (Sh0ny)
5 июля 2026
  1. Home
  2. Blog
  3. Anthropic in 5 Days: Sonnet 5 as the Default, the Return of Fable 5, and New Security Measures
1 min read

In short

Anthropic rolled out Sonnet 5 and made it the default model across all pricing plans, brought back Fable 5 after a 19-day outage, and added several cybersecurity tools. Let’s take a look at what has actually changed for users.

In just five days, Anthropic has accomplished quite a bit: a new default model, the return of the previously disabled Fable 5, and a package of safety initiatives. Claude Code users noticed that Sonnet 5 has appeared in the header without the need for manual switching—and this isn’t a bug, but a deliberate decision.

Sonnet 5 — The New Default

The model is set as the default on all plans. According to the reported data, Sonnet 5 scores 80.4% on Terminal-Bench, outperforming Opus 4.8, while costing three times less. At first glance, this represents a significant shift in the price-to-performance ratio.

However, there’s a catch: according to the author’s observations, the new tokenizer cancels out 30–40% of the claimed savings. In other words, the actual benefit from the price reduction turns out to be significantly more modest than the marketing figures suggest.

The Return of Fable 5

After 19 days “on hold”, Fable 5 is available again. Along with its return, Anthropic introduced several security mechanisms:

  • cyber safeguards — protective restrictions for cyber scenarios;
  • jailbreak severity framework — a system for assessing the severity of attempts to bypass security;
  • integration with HackerOne for bug hunters searching for vulnerabilities.

What Happens in Practice

The author tested Sonnet 5 on their work tasks and shares their observations: the model does indeed behave differently, but the savings on tokens are offset by the new tokenizer. The question of where exactly the real benefit has “moved” remains open.

Source: All Articles / Artificial Intelligence / Habr

новостиaillmбезопасность
Liked this write-up? Get one like it in your inbox every week
​

Comments

(0)
​