Anthropic has released Claude Sonnet 5, a midsize AI model that breaks the traditional trade-off between capability and cost by delivering near-flagship performance at a fraction of the price. Released on June 30, 2026, Sonnet 5 scores 63.2% on SWE-bench Pro, a demanding software-engineering benchmark that measures a model's ability to resolve real repository issues end to end, placing it within six percentage points of Claude Opus 4.8's 69.2% score. During an introductory pricing period through August 31, 2026, Sonnet 5 costs $2 per million input tokens and $10 per million output tokens, which is exactly 40% of Opus 4.8's standard rate of $5 and $25 respectively.

Why Does This Price Difference Matter So Much for AI Agents?

The pricing gap becomes transformative when you understand how autonomous AI agents work. Unlike a single question-and-answer exchange, an agent that browses the web, reads files, or writes code makes dozens or hundreds of tool calls, each generating output tokens that accumulate rapidly. Cutting the output token price from $25 to $10 per million tokens does not shave a few percentage points off a monthly bill; it fundamentally changes which autonomous workflows are economically viable to run at all. For teams building agents that run for hours or days, this cost reduction can mean the difference between a prototype and a production system.

Anthropic positioned Sonnet 5 as "built to be the most agentic Sonnet model yet," emphasizing that it "can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models". The company is backing this claim with specific performance metrics on agent-focused benchmarks. On OSWorld-Verified, a computer-use benchmark where the model operates a real desktop environment, Sonnet 5 reaches 81.2%, up from Sonnet 4.6's 78.5%. On BrowseComp, which measures autonomous web search capabilities, the model demonstrates comparable improvements.

How to Evaluate Claude Sonnet 5 for Your Use Case

• Benchmark Performance: Compare Sonnet 5's 63.2% SWE-bench Pro score against your specific task requirements; if your work demands the absolute highest performance on complex engineering problems, Opus 4.8's 69.2% score may justify its 150% price premium, but for most agentic workflows, the six-point gap is acceptable.
• Token Consumption Patterns: Calculate your typical output token usage; if your agents generate hundreds of thousands of tokens per task, the 60% discount on output pricing during the introductory period could reduce costs by tens of thousands of dollars monthly compared to flagship models.
• Safety Requirements: Evaluate whether prompt-injection resistance and lower hallucination rates matter for your deployment; Sonnet 5 shows improved performance in both areas compared to its predecessor, making it safer for agents that interact with untrusted data sources.
• Timeline for Migration: If you plan to adopt Sonnet 5, moving production traffic before August 31, 2026 locks in the introductory pricing; after that date, standard pricing rises to $3 per million input tokens and $15 per million output tokens, though this still undercuts Opus 4.8 by 40%.

The strategic move here extends beyond pricing. Anthropic made Sonnet 5 the default model for all free and Pro users on Claude.ai, meaning the majority of users who never change the model dropdown are now running an agent-capable model without opting in. This distribution strategy instantly moves Anthropic's largest audience onto a more capable foundation for autonomous workflows. Every free user is now, in effect, running a model that can plan, use tools, and execute multi-step tasks, seeding the user base for a future where chat is only the entry point and autonomous workflows are the product.

Safety considerations are particularly important when deploying agents at scale. Anthropic reports that Sonnet 5 "shows an overall lower rate of undesirable behaviors than Sonnet 4.6, and is generally safer to use in agentic contexts". Two safety improvements stand out for builders deploying agents: the model is "better at refusing malicious requests and resisting hijack attempts in prompt injection attacks," and it shows "lower rates of hallucination and sycophancy than Sonnet 4.6". Prompt-injection resistance is especially critical for agents that browse the web or read untrusted files, as these are standing targets for injected instructions hidden in external content.

Anthropic

The introductory pricing window itself reveals Anthropic's strategy. By pricing Sonnet 5 below even Sonnet 4.6 until August 31, the company is buying migration and habit formation. Teams that move production traffic onto claude-sonnet-5 now lock in the behavior before the September step-up, and the step-up itself is gentle; users land back at the familiar Sonnet 4.6 price, not above it. It is a discount designed to become a default.

Sonnet 5 is available immediately as the default model for free and Pro plans on Claude.ai, and to Max, Team, and Enterprise subscribers. Developers can access it through Claude Code and the Claude API using the model identifier claude-sonnet-5. The release represents a clear inflection point in AI economics: the unwritten rule of frontier AI for the past two years was simple, if you wanted the smartest model, you paid flagship prices, and if you wanted cheap tokens, you accepted a real drop in capability. Sonnet 5 is Anthropic's clearest attempt yet to break that rule.