Artificial Intelligence

Anthropic Launches Claude Sonnet 4.5

Published OCT 2 2025

Story

Above: The Claude Sonnet 4.5 logo displayed on a smartphone screen. Image credit: Thomas Fuller/SOPA Images/LightRocket/Getty Images

The Facts

Anthropic released Claude Sonnet 4.5 on Monday, which the company claims is the best coding model in the world, based on industry benchmarks, including software engineer benchmarks that measure AI systems' software coding abilities.
guardian
Fox2Now
Decrypt
The new model can operate autonomously for up to 30 hours straight on complex coding tasks, an increase from the previous Claude Opus 4 model, which could handle coding tasks for up to seven hours by itself.
Verge
Bloomberg.com
Claude Sonnet 4.5 achieved a 61.4% performance on key benchmarks testing real-world computer tasks, compared to the previous Sonnet 4 model, which had gained 42.2% just four months earlier.
Breitbart
Medium
During safety evaluations, Claude Sonnet 4.5 demonstrated situational awareness by recognizing when it was being tested. The model showed suspicion about testing scenarios approximately 13% of the time during automated assessments.
Anthropic
guardian
The model is available through the Claude API at the same pricing as Claude Sonnet 4, costing $3 per million input tokens and $15 per million output tokens, with no predetermined functionality or prewritten code in its operations.
Anthropic
TechCrunch
Anthropic launched the Claude Agent SDK alongside the model release, providing developers with the same infrastructure that powers Claude Code to build their own AI agents for various applications beyond just coding tasks.
TechCrunch
Anthropic

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Techno-optimist narrative

Claude Sonnet 4.5 is a revelation — an AI that doesn't just answer, but endures, thinks, and creates with unmatched precision. Mastering coding, reasoning, and autonomy, it sustains focus for days, navigates complex tasks, and sets new benchmarks with elegance. In its clarity and capability, Anthropic has crafted not just a model, but a standard for intelligent systems.

Medium

Decrypt

Techno-skeptic narrative

Anthropic's Claude Sonnet 4.5 may impress in benchmarks, but its tendency to "catch on" to being tested should cool the hype. What's framed as situational awareness is just another reminder that AI can act differently under scrutiny than in real use. Until independent trials prove otherwise, bold claims risk overselling a tool still riddled with limitations.

guardian