Anthropic launched Claude Sonnet 5, the most recent Sonnet-class mannequin. Though it’s not a frontier-model breakthrough, Sonnet 5 meaningfully upgrades efficiency over earlier fashions to ship stronger coding capabilities, higher agentic efficiency, and extra environment friendly token utilization.
Anthropic’s announcement emphasised agentic efficiency, particularly the mannequin’s capacity to hold out multi-step work with much less direct human steerage. Anthropic says Sonnet 5 could make plans, use instruments comparable to browsers and terminals, and function autonomously at a degree that lately required bigger, costlier fashions.
Sonnet 5 Is Extra Economical With Tokens
Anthropic exhibits that Sonnet 5 improves over 4.6 with lower-price choices and better high quality. Opus 4.8 nonetheless beats Sonnet 5 for accuracy, however Anthropic says that the hassle degree might be adjusted to search out one of the best stability between price and efficiency. There’s additionally an introductory worth for Sonnet 5 of $2/MTok enter and $10/MTok output via August 31.
Sonnet 5 Efficiency Benchmarks
Sonnet 5 beats Sonnet 4.6, GPT-5.5 and Gemini 3.5 Flash throughout quite a few benchmarks.
The BrowseComp assessments how nicely an AI agent can find troublesome to search out data on the internet.
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 a check of an AI mannequin’s capacity with coding duties in 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 Professional is a software program engineering benchmark during which Sonnet 5 outperformed different comparable LLMs.
SWE-bench Professional 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 a benchmark for agentic coding throughout 150 duties, a benchmark that Sonnet 5 considerably outperformed GPT-5.5.
The Claude Sonnet 5 System Card explains:
“Every job provides the agent a checked-out repository and a single challenge description; the agent then works autonomously in a containerized atmosphere to provide a ultimate patch, with no human intervention and no timeout data.
Patches are graded in opposition to blocking purposeful standards (primarily held-out unit assessments) plus weighted rubric standards, together with model-graded checks for required check protection and prohibited implementation patterns. Duties have been authored by maintainers of the underlying repositories and individually reviewed by Cognition researchers, with a random subset manually solved toverify equity.”
The FrontierCode scores:
- Claude Sonnet 5: 38.8
- Claude Sonnet 4.6: 15.1
- GPT-5.5: 25.5
Sonnet 5 Is “Close to-Opus Intelligence”
Anthropic doesn’t declare that Sonnet 5 is a frontier mannequin breakthrough, though it does say that it’s their most succesful Sonnet-class mannequin. The system card explains that it’s much less succesful than Anthropic’s extra succesful Opus and Mythos fashions. But Anthropic does declare that it’s “near-Opus intelligence at Sonnet pricing for coding, brokers, and on a regular basis skilled work.”
Learn the complete announcement at Anthropic.
Featured Picture by Shutterstock/jackpress
