Anthropic released Claude Sonnet 5 on July 1 as its most capable mid-tier model yet, priced at just $2 per million input tokens while approaching the performance of its flagship Opus 4.8 model. The launch marks a significant shift in how AI companies compete: instead of racing to build the smartest single model, they're now focusing on offering capable models at sustainable prices and teaching developers how to use them strategically.

Why Does Claude Sonnet 5's Pricing Matter So Much?

The economics of AI-powered products hinge on model cost. At $2 per million input tokens, Sonnet 5 makes previously expensive use cases viable. A chatbot that once required expensive API calls becomes profitable at scale. Document processors that previously needed smaller, less capable models can now use Sonnet's superior reasoning without breaking budgets.

This pricing strategy reflects a broader industry trend. OpenAI released GPT-5.6 on July 9 with multiple pricing tiers, and both companies are pursuing similar three-tier structures: a capable mid-range model, a frontier model for complex tasks, and a small, fast, cheap option for simple work. The standardization suggests the industry has moved past the "one model fits all" era.

What's the Real Difference Between Models and "Effort"?

A major revelation in the Claude ecosystem came when Anthropic explained why users experienced what felt like sudden degradation in Claude's abilities earlier this year. In March 2026, developers reported that Claude Code had become noticeably "dumber," with AMD's AI chief measuring a 67% plunge in reasoning output. The culprit wasn't model degradation at all.

Anthropic had quietly lowered the default "Effort" setting from High to Medium on March 4 to reduce response latency. This single toggle determines how much computational work Claude is willing to invest in a task, separate from the model's underlying capabilities. High Effort generates roughly 7 times more tokens than Low Effort, spending those extra resources on reading files, running tests, and verifying results.

"The Model changes the brain; Effort changes the attitude," Anthropic explained in its official technical documentation.

Anthropic, Official Technical Blog

This distinction has profound implications. A smaller model like Sonnet paired with High Effort can outperform a larger model like Opus running on Low Effort for specific tasks. The company restored the default to High on April 7 and reset usage quotas for all paying users after enduring nearly a month of public pressure.

How to Optimize Your AI Model Selection and Effort Settings

• Simple Tasks on Low Effort: Use Sonnet with Low Effort for quick code changes and straightforward requests, achieving sub-second responses while controlling costs.
• Complex Work on High Effort: Deploy stronger models with High Effort for major refactoring, multi-step workflows, and tasks requiring thorough verification and file reading.
• Extended Autonomous Operations: Equip agent tasks that require extended independent operation with ample computational firepower, using Anthropic's "ultracode" tier for tasks that benefit from parallel decomposition.

Anthropic has painted clear personas for its different models to help developers make these choices. Sonnet is "the all-rounder with an entire afternoon" that will read code from start to finish, execute tests, and thoroughly digest tasks. Opus is "the expert who only gives you five minutes," striking at the heart of problems based on experience but lacking time to scan every file. Fable, the most expensive model, is "the expert you only call in when everyone else is stuck," reserved for the toughest challenges.

What Does This Mean for the AI Competition Landscape?

The "brain fog" episode revealed a fundamental shift in how AI companies compete. The race is moving from "whose model is stronger" to "who is better at orchestrating agents." For developers and enterprise users, learning how to precisely assign work to AI by combining different models and Effort levels is becoming the core skill that determines productivity.

OpenAI is pursuing the same strategy. Its newly released GPT-5.6 model and ChatGPT Work feature directly target Claude Cowork, Anthropic's agent-powered workspace tool. In real-world task tests, GPT-5.6 Sol delivered results with higher completion quality and could generate publicly shareable web links. However, OpenAI's ChatGPT Work suffers from rapid Agent quota consumption, with billing hidden outside the subscription.

Anthropic's Claude Cowork retains unique strengths in desktop workflows such as handling local files and organizing meeting materials. In a competitive analysis that ChatGPT Work generated about itself, it objectively noted that "Cowork is more mature; Work is broader," recommending that production-oriented pilot programs prioritize Cowork.

The timing of these releases suggests strategic positioning. Anthropic released Sonnet 5 on July 1, before OpenAI's July 9 release of GPT-5.6, establishing market position ahead of the competition. However, first-mover advantage in AI model pricing is real but fleeting; feature velocity and orchestration capability matter more in the long run.

Sonnet 5 isn't revolutionary in isolation. It's iterative, approaching Opus performance while costing significantly less. That iteration at lower cost to the customer is the real story. As Anthropic continues to iterate Sonnet and Opus while improving Haiku, the pace of change has accelerated dramatically. GPT-4.5 to GPT-5 took roughly a year; GPT-5 to GPT-5.6 variants took just two months.

For customers, the benefits are clear: faster iteration cycles, lower costs, and models available immediately without waitlists or approval processes. Sonnet 5 is available on Claude.ai and the Claude Platform API, signaling Anthropic's confidence in the model's quality and safety. The era of judging AI capability by model leaderboards alone is passing. The new era belongs to those who master the art of assigning work intelligently across a portfolio of models and settings.