Anthropic announced Claude Sonnet 5, its newest AI model, on June 30, 20261. The latest entry in the mid-tier Sonnet series delivers a substantial improvement over its predecessor Sonnet 4.6 on the core aspects of agentic performance — reasoning, tool use, coding, and knowledge work — bringing its capabilities close to the higher-end Opus 4.8 while keeping prices low1.
Availability began on launch day: Sonnet 5 is the default model for Free and Pro plans, and is also available to Max, Team, and Enterprise users, in Claude Code, and on the Claude Platform1. Developers can call it via the Claude API using the model ID claude-sonnet-51. Because it replaces the standard model for a broad user base — including free users — this is a release with a wide footprint.
The Most Agentic Sonnet Yet
According to Anthropic, Sonnet 5 was built to be “the most agentic Sonnet model yet.” It can make plans, use tools like browsers and terminals, and run autonomously1. The design emphasizes practical utility as an AI agent.
On an agentic coding benchmark, Sonnet 5 scored 63.2%. That falls short of Opus 4.8’s 69.2%, but is a significant jump from Sonnet 4.6’s 58.1%2. In knowledge work, it reportedly even slightly outperforms Opus 4.82, and Anthropic itself says the model “in some cases matches Opus 4.8’s capability levels”1.
Early access partners reported noticeable behavioral changes: the model finishes complex tasks where previous versions would have stopped short, and checks its own output without being explicitly asked2.
Users can also adjust the effort level to find the right balance of cost and performance1. Effort settings include tiers such as medium and xhigh (extra high), so light tasks can run cheaply while harder problems get more reasoning power — all within a single model. On the technical side, Sonnet 5 uses an updated tokenizer that changes how the model processes text to improve performance1. Note, however, that the new tokenizer produces approximately 30% more tokens for the same text, so API cost estimates should account for the higher token counts, not just the per-token price3.
Pricing - Introductory Rates Through the End of August
The API launches with limited-time introductory pricing: $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which standard pricing of $3 input / $15 output takes effect1.
TechCrunch notes that this pricing undercuts not only Opus 4.8 but also OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro2. Since API costs are often the practical barrier to running agents at scale or for long stretches, the “near-Opus performance at a lower price” positioning looks like a meaningful option for reducing the cost of agent operations.
Safety - Fewer Undesirable Behaviors, Cyber Safeguards On by Default
On the safety side, Sonnet 5 shows an overall lower rate of undesirable behaviors than Sonnet 4.61. Specifically, behaviors like cooperation with misuse and deception are reduced, and the model is better at refusing malicious requests and resisting instruction hijacking such as prompt injection2. Hallucination — generating plausible-sounding misinformation — and sycophantic behavior are also down compared to Sonnet 4.62.
Sonnet 5 also launched with cyber safeguards enabled by default1. Since the Claude Fable 5 and Mythos 5 announcement in June, Anthropic has been building mechanisms to reconcile rising model capability with misuse prevention, and this move follows that same trajectory.
Raising the Baseline
Two aspects of this release stand out. First, the standard model used by the vast majority of Claude users — including free users — gets a performance upgrade. Because the default for Free and Pro plans switches to Sonnet 5, even users who never consciously pick a model will move to an environment with stronger agentic performance1.
Second is the impact on the cost structure of running AI agents. Until now, agent tasks requiring high reliability tended to call for top-tier models. With a “near-Opus 4.8 performance at a fraction of the price” option now available, the barrier to trying coding assistance or workflow automation should come down. The introductory pricing window through the end of August makes it an even easier time to experiment.
If you are deciding which AI coding tool to try it with, see our AI coding tools comparison guide.
Sources
- Introducing Claude Sonnet 5 - Anthropic official announcement
- Anthropic launches Claude Sonnet 5 as a cheaper way to run agents - TechCrunch
- Pricing - Claude Docs - Anthropic official documentation