Anthropic released Claude Sonnet 5 on June 30, 2026. According to Anthropic, the new mid-range model performs at a level close to that of the flagship model, Opus 4.8, but costs significantly less and is the most agent-like Sonnet model to date out of the box. It replaces Sonnet 4.6 and is now available in all plans, in Claude Code, and via the API.
Key takeaways:
- Claude Sonnet 5 replaces Sonnet 4.6 and is now the default model for Free and Pro users.
- Anthropic calls it the most agent-like Sonnet yet: It plans tasks, uses browsers and terminals, and works more independently.
- Its performance is close to that of Opus 4.8; in some tests involving knowledge work, Sonnet 5 even narrowly outperforms it.
- Introductory price: $2 per million input tokens, $10 per million output tokens—through August 31, 2026.
- After that, the price will rise to $3 and $15 per million tokens, respectively.
What Makes Claude Sonnet 5 Better
For many developers, Sonnet models were their first foray into agentic AI. Even Sonnet 3.5, 3.6, and 3.7 demonstrated strong coding and tooling capabilities. Recently, however, the biggest leaps have come from the more expensive Opus series. Anthropic aims to bridge precisely this gap with Claude Sonnet 5.
Compared to its predecessor, Sonnet 4.6, the model has improved in the areas that matter most in everyday use: reasoning, tool usage, programming, and knowledge work. According to Anthropic, test partners report that Sonnet 5 completes complex tasks that earlier Sonnet versions abandoned, while verifying its own results without being explicitly asked to do so. For example, a Zapier engineer describes how the model fully executed a two-part task—updating Salesforce accounts and sending an announcement to enterprise contacts—while earlier models stalled halfway through.
Claude Sonnet 5 vs. Sonnet 4.6 and Opus 4.8
The most interesting point is how it compares to Opus 4.8. In terms of agent-based programming, Sonnet 5 still lags behind the top-of-the-line model, but it covers a significantly broader range of cost-performance options. Using the adjustable effort level, the model can be fine-tuned to favor speed and cost-effectiveness or a slower, more thorough approach. For some types of knowledge work, Sonnet 5 even performs slightly better than Opus 4.8.
| Feature | Sonnet 4.6 | Claude Sonnet 5 | Opus 4.8 |
|---|---|---|---|
| Agent-Based Coding (Benchmark) | 58.1% | 63.2% | 69.2% |
| Input Price (per million tokens) | $3 | $2 (launch) / $3 | $5 |
| Output price (per million tokens) | $15 | $10 (launch) / $15 | $25 |
| Hallucinations / Sycophancy | higher | lower | lowest |
| Cyber Skills | low | low (with safeguards) | higher |
For context: Anthropic had just recently introduced Claude Opus 4.8, which remains the top choice for maximum accuracy and complex reasoning tasks. Sonnet 5 is the more affordable alternative for the majority of daily tasks. If you’re familiar with its direct predecessor, our article on Claude Sonnet 4.6 provides the foundation upon which the new model is built.
Pricing and Availability
Claude Sonnet 5 is now available everywhere. At launch, an introductory price of $2 per million input tokens and $10 per million output tokens applies through August 31, 2026. After that, the model will switch to $3 (input) and $15 (output) per million tokens. This makes Sonnet 5 more affordable than Opus 4.8, as well as OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro.
There’s one technical detail you should keep in mind: Sonnet 5 uses a new tokenizer. As a result, the same text can be represented by 1.0 to 1.35 times as many tokens, depending on its content. Anthropic has set the introductory price so that switching from Sonnet 4.6 is roughly cost-neutral. You can use the model via claude.ai (web, iOS, Android), in Claude Code, Cowork, and via the API with the identifier claude-sonnet-5. For Free and Pro users, it’s the new default model.
Security: More Secure Than 4.6, but Below Opus Level
In security tests, Sonnet 5 performs better overall than Sonnet 4.6. The model more reliably rejects malicious requests, better defends against prompt injection attempts, and exhibits lower rates of hallucinations and excessive agreement (sycophancy). However, when it comes to erroneous behavior, it still exceeds the values of the stronger models Opus 4.8 and Claude Mythos Preview.
Sonnet 5 is intentionally weaker when it comes to cybersecurity tasks. In a test developed jointly with Mozilla to create exploits for Firefox vulnerabilities, the model did not produce a single working exploit. Nevertheless, Sonnet 5 comes with cyber safeguards enabled by default that detect and block dangerous usage in real time. For serious cybersecurity work, Anthropic continues to recommend Opus 4.8. Our overview of Claude Fable 5 and Mythos 5 shows how the model family is otherwise tiered.
Frequently Asked Questions About Claude Sonnet 5
When was Claude Sonnet 5 released?
Anthropic released Claude Sonnet 5 on June 30, 2026, about four months after Sonnet 4.6.
How much does Claude Sonnet 5 cost?
Until August 31, 2026, an introductory price of $2 per million input tokens and $10 per million output tokens applies. After that, the price will increase to $3 and $15 per million tokens, respectively. For Free and Pro users of Claude, this is the default model at no extra charge.
Is Claude Sonnet 5 better than Opus 4.8?
Not across the board. For agent-based programming, Opus 4.8 comes out ahead (69.2% vs. 63.2%). For some knowledge work tasks, Sonnet 5 narrowly takes the lead. For the highest accuracy, Opus 4.8 remains the recommended choice; for the best value for money, Sonnet 5 is the top pick.
How do I use Claude Sonnet 5?
Via claude.ai in the browser, as well as in the iOS and Android apps, in Claude Code and Cowork, and via the API using the model ID claude-sonnet-5. It’s also available on AWS, Google Cloud, and Microsoft Foundry.
Conclusion
Claude Sonnet 5 does what a good mid-range model is supposed to do: It delivers much of Opus’s power at a lower price while becoming noticeably more independent. Those who previously had to choose between the expensive Opus and the less powerful Sonnet 4.6 now have finer control over costs and quality thanks to the Effort slider. For everyday AI tasks—coding, automation, and agent-based workflows—Sonnet 5 is likely to be the new go-to choice for many. Only for truly challenging reasoning and cybersecurity tasks does Opus 4.8 remain the top choice.
Source: Anthropic
