Editorial illustration for Claude Sonnet 5 Launches With User-Adjustable Effort Level
Claude Sonnet 5 Launches With Adjustable Effort
Anthropic released Claude Sonnet 5 today, calling it the most agentic Sonnet model the company has built. The model plans multi-step tasks, operates browsers and terminals, and runs on its own at a level that, a few months back, only larger and pricier models could manage.
Sonnet-class models have carried the agentic AI push before. Claude Sonnet 3.5, 3.6, and 3.7 were the first to show real skill at coding and tool use, setting the template developers built on. Since then, the biggest jumps in agentic performance had shifted to Anthropic's Opus line, leaving Sonnet models a step behind. Sonnet 5 is meant to close that distance: Anthropic says its performance now sits close to Opus 4.8, at a fraction of the cost, and marks a clear jump over its predecessor, Sonnet 4.6, on reasoning, tool use, coding, and knowledge work.
Sonnet 5 is now the default model for Free and Pro users, and it's available to Max, Team, and Enterprise plans too. API pricing runs $2 per million input tokens and $10 per million output tokens, accessible through claude-sonnet-5. Anthropic also ran safety evaluations on the new model, comparing its behavior against Sonnet 4.6 and its current Opus models.
Sonnet 5 narrows the gap: its performance is close to that of Opus 4.8, but at lower prices. It’s a substantial improvement over its predecessor, Sonnet 4.6, on important aspects of agentic performance like reasoning, tool use, coding, and knowledge work:
Why this matters
The effort-level dial matters more than the benchmark bragging. For years, choosing a Claude model meant picking a fixed tradeoff between cost and capability upfront. Now Anthropic is letting developers tune that tradeoff per task, which is a more honest admission that "agentic" isn't one setting but a spectrum. Cheap monitoring tasks don't need the same reasoning budget as a multi-step deployment pipeline, and pretending otherwise wastes money at scale.
For founders building on Claude, this is a pricing lever disguised as a feature. Watch how usage patterns shift once teams can throttle effort instead of switching models entirely. That's a signal about where the real cost pressure in agentic AI actually sits.
For researchers, the interesting claim is that Sonnet 5 closes ground Opus opened. If early access testers are right that it finishes tasks previous Sonnet versions couldn't touch, that's a data point on how fast capability trickles down tiers, not just up them. Worth checking whether that holds outside Anthropic's own testers once independent benchmarks land.
Related Reading
Browse all Anthropic news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic Trump cracks down on Anthropic after Amazon tip; staff largely foreign
Open Source
Related Topic Claude Mythos highlights EU AI safety gaps, says researcher Caroli
Policy & Regulation
From Archives Anthropic Watermarks AI Output, Some Clients Request It
From Archives Leading mathematicians call AI strong at calculation, weak at creative thought
From Archives Anthropic Details How Claude's New AI Watermarks Will Work
Policy & Regulation
From Archives GLM-5.3 Scores 66.9 on DeepSWE v1.1, Trails Behind GPT-5 and Claude
Open Source
Common Questions Answered
How does Claude Sonnet 5's performance compare to Opus 4.8?
Claude Sonnet 5 narrows the performance gap significantly, delivering capabilities close to Opus 4.8 while maintaining lower prices. This represents a substantial improvement over its predecessor, Sonnet 4.6, particularly in agentic performance areas like reasoning, tool use, coding, and knowledge work.
What is the user-adjustable effort level feature in Claude Sonnet 5?
The effort-level dial allows developers to tune the tradeoff between cost and capability on a per-task basis, rather than being locked into a fixed model choice. This means cheap monitoring tasks can use lower reasoning budgets while complex multi-step deployment pipelines can access higher computational resources, optimizing costs at scale.
What agentic capabilities does Claude Sonnet 5 demonstrate?
Claude Sonnet 5 can plan multi-step tasks, operate browsers and terminals, and run autonomously at a level previously only available in larger and more expensive models. The model shows particular strength in reasoning, tool use, coding, and knowledge work compared to earlier Sonnet versions.
How did earlier Sonnet models contribute to agentic AI development?
Claude Sonnet 3.5, 3.6, and 3.7 were the first to demonstrate real skill at coding and tool use, establishing the foundational template that developers have since built upon. These models set the standard for what agentic capabilities should look like in the Sonnet-class lineup.
Why is the effort-level feature more practical than fixed model selection?
The effort-level dial acknowledges that 'agentic' performance exists on a spectrum rather than as a single fixed setting, allowing developers to match computational resources to task complexity. This prevents wasting money on expensive reasoning budgets for simple tasks while still providing sufficient capability for demanding multi-step operations.
Further Reading
- Introducing Claude Sonnet 5 - Anthropic
- What's new in Claude Sonnet 5 - Claude Platform Docs
- Prompting Claude Sonnet 5 - Claude Platform Docs
- Introducing Claude Sonnet 5 on AWS: Anthropic's most capable Sonnet model - AWS Machine Learning Blog
- Claude Sonnet 5: Benchmarks Explained - Vellum
