Anthropic has launched Claude Sonnet 5.5
Artificial Analysis ● Covered by 12 sources
Anthropic launched Claude Sonnet 5.5, and at full effort it jumps to second on the intelligence index. It’s fast on terminal tasks, but it burns far more tokens to get there.
Based on reporting by Artificial Analysis — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Anthropic has launched Claude Sonnet 5.5, and the pitch is simple: keep Sonnet pricing, push the model much harder. On the Artificial Analysis Intelligence Index, it lands at 56, just 2 points behind Opus 5.5 max. With max effort, it picks up 18 points over Sonnet 5 and moves into second place on the leaderboard.
The price tag has not moved. Anthropic is charging the same rates as Sonnet 5: $0.2 per 1M cache input tokens, $2 per 1M input tokens, and $10 per 1M output tokens. But that sameness hides the catch. Sonnet 5.5 produces more output tokens per task than before, and the cost per task comes out to $7.60, about 50% higher than Sonnet 5.
This is where the model gets interesting. On Terminal-Bench 4.0, it scores 64%, ahead of the 60% posted by Opus 5.5 and GPT-6 Astra. It also reaches parity with Opus 5.5 on AA-Briefcase, GDPval-AA, and AutomationBench-AA, though the source makes clear that it does so with significantly higher token use.
That token use is the real story. At max effort, Sonnet 5.5 uses about 193,000 output tokens per Intelligence Index task, the most measured so far, around 60% more than Opus 5.5 max or Sonnet 5 max, and roughly 7 times GPT-6 Astra max. Lower effort settings are more mixed, and on the Intelligence-versus-cost tradeoff they sit behind GPT-6 Sol configurations that deliver similar or better performance for less.
Anthropic’s own pre-release testing also found a bug that could hurt responses to structured-output prompts. The company says that issue is fixed for the public release, and expects performance to stay roughly the same or inch up a bit. Sonnet 5.5 keeps the same 1 million token context window as Sonnet 5, so this is less a new product shape than a more expensive, more aggressive one.
My take — AI-written commentary, not fact-checked reporting
This is classic frontier-model behavior now: the bragging rights come from making the model sweat harder, then calling the token bill a feature. Sonnet 5.5 looks strong, but the market should stop pretending “same price” means the same cost when task usage is doing all the talking. The token meters are the new product launch notes.
Read more about this at: Artificial Analysis
