Anthropic is shaking up its artificial intelligence family with a major mid-tier upgrade. The company officially released Claude Sonnet 5.5 across major cloud hubs including AWS, Google Cloud, and Microsoft Azure using the model ID “claude-sonnet-5-5,” alongside zero data retention policies.
This new release generates output over 30% faster while cutting effective per-task costs by up to 30%. Anthropic positioned Sonnet 5.5 for well-defined daily tasks like fixing bugs, writing documentation, building presentation decks, and editing complex spreadsheets. The firm also teased that a smaller, high-throughput Haiku 5.5 build will follow in the coming weeks.
Massive jumps in agentic coding performance
The performance leap in coding benchmarks is where Claude Sonnet 5.5 stands out most. On the Terminal-Bench 4.0 agentic coding evaluation, Sonnet 5.5 hit an impressive 70.6%, leaping past Sonnet 5’s 10.3% and beating the flagship Opus 5.5, which scored 66.4% at its highest effort setting.
On CursorBench 4.0, which recreates real coding sessions, Sonnet 5.5 reached 55.5% against Sonnet 5’s 34.1%, landing just behind Opus 5.5 at 57.8%. For FrontierCode 1.1 at High effort, it scored ten points above Sonnet 5 at one-fifteenth the task cost. The AI tech lab also noted an odd wrinkle where the “Max” effort level triggered extra sub-agent code reviews that led to timeouts.
Knowledge work matching flagship tiers
Beyond coding, Sonnet 5.5 delivers flagship-level results across general professional benchmarks. On OpenAI’s GDPval-AA knowledge test across 44 occupations, Sonnet 5.5 scored 1,844 points, virtually matching Opus 5.5 at 1,846 points and far outpacing OpenAI’s GPT-6 Sol at 1,487 points.
Visual recognition saw massive gains too, with Chartography scores jumping from 15.6% to 61.6%. Early testers praise how naturally the model formats user interfaces and slide templates with minimal edits. Meanwhile, Anthropic even highlighted that Sonnet 5.5 can play through Pokémon Red using only raw screenshots.
Pricing structure and frontier cybersecurity limits
Token pricing stays at $2 per million input tokens, $10 per million output tokens, and $0.20 for cache reads. Effective costs drop because Sonnet 5.5 batches tool calls more efficiently. In other words, the model requires far fewer execution steps per prompt.
Because its technical reasoning jumped so sharply, Anthropic deployed frontier-style cybersecurity safeguards to a Sonnet build for the first time. High-risk cyber queries visibly fall back to Sonnet 5, though vetted experts can apply for tiered access through an expanded Cyber Verification Program.
Lastly, the model introduces classifiers to block distillation attacks and reasoning extraction. This protection follows recent research that decoded 315,320 public agent thinking blocks, helping Anthropic comply with European Union AI Act rules regarding systemic risk protection.
The post Claude Sonnet 5.5 Arrives with Flagship Coding Power at Half the Price of Opus appeared first on Android Headlines.
