Claude Unveils a Tech Surprise
Anthropic has just launched their Opus 5 AI model, asserting it’s quite the tech marvel, closely rivaling Claude Fable 5’s intelligence but at a fraction of the cost. That’s akin to saying you’ve got Einstein’s intellect while paying just for a bag of beans. Companies are likely beaming with joy, especially with token prices skyrocketing recently.
Savvy Spending with Tokens
Alright, let’s break down the figures. Fable 5 demands $10 / MTok for input and $50 / MTok for output, while our sparkling Opus 5 is more user-friendly at $5 / MTok and $25 / MTok. Even OpenAI’s GPT-5.6 Sol isn’t too shabby at $5 / MTok for input and $30 / MTok for output. But keep in mind, various tasks consume different amounts of tokens based on the model, so check your balance before you make a purchase.
The True Expense of Intelligence
If you trust Artificial Analysis, the cost per Intelligence Index task stands at $2.75 for Fable, $2.03 for Opus 5, peaking at $1.04 for GPT-5.6 Sol, and a cheeky $0.95 for Kimi K3. Opus 5 is currently at the forefront on the Intelligence Index with a score of 61, just a single point ahead of Fable.
A Leaner, Sleeker Genius
Claude Code engineer Thariq Shihipar reveals they’ve slimmed down Opus 5 significantly, cutting 80% of the system prompt bulk. You might want to do the same with your system prompts, skills, and CLAUDE.md files, akin to transitioning from a heavy meal to a lighter option. A command named “claude doctor” can help tidy up your setup.
Chatty Cathy Syndrome
Be cautious, Opus 5 might be more talkative compared to earlier versions. Anthropic cautions that “Claude Opus 5’s standard user-facing replies are longer than those of previous models.” If you prefer brevity, you’ll need to specify that in the prompt.
Techies on a Quest
Opus 5 is nearly on par with Mythos when it comes to detecting vulnerabilities in open-source code, achieving 80% on the OSS-Fuzz benchmark compared to Mythos’s 79.4%. However, in the tricky area of weaponizing those discoveries, it scores a meager 4 out of 14 attempts versus Mythos’s 13 out of 14.
Staying Legit
Opus 5 excels in cybersecurity and biology tasks, behaving responsibly and not jumping at the chance for mischief. Its cyber classifiers are less overbearing than those found in Fable 5, identifying vulnerabilities while blocking any dubious activities such as “binary-based” vulnerability scanning, penetration testing, and exploit generation.
Backup Strategy for the Win
Anthropic is introducing a backup feature on its API, so when Opus 5 or Fable 5 decline to cooperate, your request will be routed to a less demanding model rather than being completely denied.
Opus 5 – The Aligned Partner
Anthropic boasts that Opus 5 is their most “aligned” model, the top student less likely to stray off course. The major benefit for businesses? No data retention issues to worry about here.
Summary: A Steal or a Chatterbox?
There you have it, everyone. Opus 5 offers intelligence and an affordable price tag, but it might be a bit chatty. Still, it could represent the AI deal of the year. Explore more about it on GadgetLad if you’re interested!
