Anthropic released Claude Sonnet 5, the latest Sonnet-class model. Although it’s not a frontier-model breakthrough, Sonnet 5 meaningfully upgrades capacity complete erstwhile models to present stronger coding capabilities, amended agentic performance, and much businesslike token usage.
Anthropic’s announcement emphasized agentic performance, specifically the model’s expertise to transportation retired multi-step activity pinch little nonstop quality guidance. Anthropic says Sonnet 5 tin make plans, usage devices specified arsenic browsers and terminals, and run autonomously astatine a level that precocious required larger, much costly models.
Sonnet 5 Is More Economical With Tokens
Anthropic shows that Sonnet 5 improves complete 4.6 pinch lower-price options and higher quality. Opus 4.8 still thumps Sonnet 5 for accuracy, but Anthropic says that the effort level tin beryllium adjusted to find the champion equilibrium betwixt costs and performance. There is besides an introductory value for Sonnet 5 of $2/MTok input and $10/MTok output done August 31.
Sonnet 5 Performance Benchmarks
Sonnet 5 thumps Sonnet 4.6, GPT-5.5 and Gemini 3.5 Flash crossed a number of benchmarks.
The BrowseComp tests really good an AI supplier tin find difficult to find accusation connected the web.
BrowseComp scores:
- Claude Sonnet 5: 84.7 (single agent)
- Claude Sonnet 4.6: 76.2
- GPT-5.5: 84.4
Terminal-Bench 2.1 is simply a trial of an AI model’s expertise pinch coding tasks successful terminal and CLI.
Terminal-Bench 2.1 scores:
- Claude Sonnet 5: 80.4
- Claude Sonnet 4.6: 67.0
- GPT-5.5: 83.4 (Codex CLI)
- Gemini 3.5: Flash 76.2
SWE-bench Pro is simply a package engineering benchmark successful which Sonnet 5 outperformed different akin LLMs.
SWE-bench Pro scores:
- Claude Sonnet 5: 63.2
- Claude Sonnet 4.6: 58.1
- GPT-5.5: 58.6
- Gemini 3.5 Flash: 55.1
FrontierCode is simply a benchmark for agentic coding crossed 150 tasks, a benchmark that Sonnet 5 importantly outperformed GPT-5.5.
The Claude Sonnet 5 System Card explains:
“Each task gives the supplier a checked-out repository and a azygous rumor description; the supplier past useful autonomously successful a containerized situation to nutrient a last patch, pinch nary quality involution and nary timeout information.
Patches are graded against blocking functional criteria (primarily held-out portion tests) positive weighted rubric criteria, including model-graded checks for required trial sum and prohibited implementation patterns. Tasks were authored by maintainers of the underlying repositories and individually reviewed by Cognition researchers, pinch a random subset manually solved toverify fairness.”
The FrontierCode scores:
- Claude Sonnet 5: 38.8
- Claude Sonnet 4.6: 15.1
- GPT-5.5: 25.5
Sonnet 5 Is “Near-Opus Intelligence”
Anthropic does not declare that Sonnet 5 is simply a frontier exemplary breakthrough, though it does opportunity that it’s their astir tin Sonnet-class model. The strategy paper explains that it is little tin than Anthropic’s much tin Opus and Mythos models. Yet Anthropic does declare that it is “near-Opus intelligence astatine Sonnet pricing for coding, agents, and mundane master work.”
Read the afloat announcement astatine Anthropic.
Featured Image by Shutterstock/jackpress
English (US) · 
Indonesian (ID) ·