Claude Sonnet 5 is Anthropic's newest mid-tier model, released June 30, 2026. It is the Sonnet-class member of the Claude 5 family, and the pitch is simple: it closes most of the gap to Anthropic's far pricier Opus class on coding and agentic work, while keeping Sonnet-level pricing. It matters because for the first time a Sonnet model finishes complex, multi-step tasks that previously needed Opus.
Yes. Anthropic released Claude Sonnet 5 on June 30, 2026. It is live now, not a preview or a waitlist: it is already the default model for Free and Pro users on Claude.ai, it is available to Max, Team, and Enterprise plans, and developers can call it through the Claude API today using the model ID claude-sonnet-5.
If you want the primary source, the full launch details live in Anthropic's official announcement.
Claude Sonnet 5 is the mid-tier model in Anthropic's Claude 5 generation. In Anthropic's lineup the Sonnet class has always been the balanced option - faster and cheaper than the flagship Opus class, but strong enough for real production work. Sonnet 5 keeps that positioning and pushes the ceiling up: Anthropic says its performance lands close to Claude Opus 4.8, the current Opus flagship, across reasoning, tool use, coding, and knowledge work.
The practical framing from early users is that Sonnet 5 finishes complex tasks where previous Sonnet models would stop short. On long, multi-step jobs it keeps going and verifies its own output without being explicitly told to, rather than handing back a half-done answer. That is the behavior that used to force teams onto Opus, and it is the main reason Sonnet 5 is interesting.
Anthropic positions Sonnet 5 as narrowing the gap to the Opus-class models while staying much more affordable. Its own description is that Sonnet 5 performs close to Opus 4.8 - not that it matches or beats it. So the honest summary is: Opus 4.8 remains the more capable model on the hardest problems, and Sonnet 5 gets most of the way there for a fraction of the cost.
Anthropic did not publish exact benchmark scores in the launch announcement, and the third-party numbers circulating for Sonnet 5 vary widely by testing harness, so we are not going to quote a single figure as fact. What Anthropic does claim directly is a substantial improvement over Sonnet 4.6 and performance close to Opus 4.8 on coding and agentic tasks. Treat any precise SWE-bench percentage you see elsewhere as an unverified third-party estimate until independent benchmarks settle.
The decision rule for most teams: default to Sonnet 5 and only escalate to Opus 4.8 when you can measure that a specific workload needs it. Sonnet 5 is cheaper to run, fast enough for interactive use, and now capable enough that the escalation happens far less often than it did with Sonnet 4.6. One caveat worth knowing: Anthropic notes Sonnet 5 has substantially weaker cybersecurity capabilities than the Opus models, with cyber safeguards enabled by default.
Claude Sonnet 5 API pricing has an introductory window and a standard rate. Through August 31, 2026, it is $2 per million input tokens and $10 per million output tokens. After that promotional period, standard pricing is $3 per million input tokens and $15 per million output tokens - the same rates Sonnet 4.6 charged, so Sonnet 5 delivers a capability jump at no long-term price increase over the previous Sonnet.
| Rate | Input | Output | When |
|---|---|---|---|
| Introductory | $2 | $10 | Through Aug 31, 2026 |
| Standard | $3 | $15 | From Sep 1, 2026 |
For comparison, Claude Opus 4.8 costs $5 per million input tokens and $25 per million output tokens. So even at Sonnet 5's standard rate, you are paying roughly 40 percent of the Opus input price and 60 percent of the output price for a model Anthropic says performs close to Opus. That cost-to-capability ratio is the whole story of this release.
The API model ID is claude-sonnet-5. Sonnet 5 is the default model for Free and Pro plans on Claude.ai, and it is available to Max, Team, and Enterprise users. Developers reach it through the Claude API. It carries the Claude 5 family's large context window - 1,000,000 tokens - with a maximum output of 128,000 tokens per request.
On behavior, Sonnet 5 makes plans and uses tools autonomously, driving browsers and terminals to complete work rather than just suggesting steps. Anthropic highlights agentic search and computer use at cost-effective price points as a core use case, which is what the near-Opus quality plus Sonnet pricing unlocks.
Beyond raw capability, Anthropic frames Sonnet 5 as more trustworthy than Sonnet 4.6. It is better at refusing malicious requests and at resisting prompt-injection attacks, and it shows lower rates of hallucination and sycophancy - the tendency to tell you what you want to hear rather than what is correct. On automated safety audits it registered lower rates of misaligned behavior overall.
The one area to flag is cybersecurity. Anthropic says Sonnet 5's cyber capabilities are substantially weaker than the Opus models', and it ships with cyber safeguards enabled by default. For general product, content, and coding work that is a non-issue; if your workload touches security tooling specifically, plan around it.
For most builders the move is straightforward: make Sonnet 5 your default and reserve Opus 4.8 for the jobs you can prove need it. Coming from Sonnet 4.6, it is a clear upgrade at the same standard price, and the introductory window makes it cheaper still until September. Coming from Opus 4.8, run your own eval on a real workload before switching - Anthropic says close to Opus, not equal to it, and the right call depends on how hard your specific tasks are and how much the cost difference matters at your volume.