Anthropic released Claude Sonnet 5 on June 30, 2026, and the one-line summary is this: it is the most agentic Sonnet yet, it performs close to the far pricier Opus 4.8 on many tasks, and it costs the same as Sonnet 4.6. It is already the default model for Free and Pro users, it is available to Max, Team, and Enterprise, and it runs in Claude Code and on the Claude Platform as claude-sonnet-5. If you use Claude at all, you are likely already talking to Sonnet 5 without changing a thing.
The interesting part is not a single benchmark, it is the positioning. For a while the biggest agentic gains lived in the expensive Opus tier. Sonnet 5 narrows that gap, bringing planning, tool use, and autonomous multi-step work down to mid-tier pricing. This guide covers exactly what changed, the real pricing including the launch discount, the safety improvements Anthropic reported, and the honest question of whether you should switch or stay on Opus.
Quick Answer: What's New in Sonnet 5
The headline changes, in plain terms, with the facts Anthropic actually published rather than the inflated numbers floating around rumor sites.
| What | Sonnet 5 | Why it matters |
|---|---|---|
| Positioning | Most agentic Sonnet yet; near Opus 4.8 on many tasks | Opus-level work at Sonnet pricing |
| Price (standard) | $3 input / $15 output per 1M tokens | Same as Sonnet 4.6, no hike |
| Launch pricing | $2 / $10 per 1M through Aug 31, 2026 | Cheaper than standard during the intro window |
| Availability | Default for Free and Pro; Max, Team, Enterprise, API | You may already be using it |
| Safety | Lower hallucination and sycophancy than Sonnet 4.6 | Fewer confident wrong answers, less flattery |
The Real Story: Agentic Work Comes Down a Tier
Anthropic frames Sonnet 5 as built to be its most agentic Sonnet model. In practice that means it can make plans, use tools like browsers and terminals, and run autonomously through multi-step tasks at a level that, until recently, needed larger and more expensive models. Earlier Sonnet versions, 3.5, 3.6, and 3.7, were the models that first showed strong coding and tool use, but lately the clearest agentic gains had been in the Opus tier. Sonnet 5 closes much of that distance, so a lot of work that used to justify paying for Opus now runs well on Sonnet pricing.
It is also a clear step up from its predecessor, Sonnet 4.6, on reasoning, tool use, coding, and general knowledge work. Anthropic shows it as a strict improvement over Sonnet 4.6 on agentic search and computer-use evaluations, while noting that Opus 4.8 still leads when you need the highest accuracy. Between the two, you can dial an effort level up or down to trade cost against precision, which is the practical way to decide which one a given task deserves.
Pricing: Same Sticker, Plus a Launch Discount
Sonnet 5 lands at the same standard price as Sonnet 4.6: $3 per million input tokens and $15 per million output tokens. Through August 31, 2026, there is an introductory rate of $2 per million input and $10 per million output, so it is actually cheaper than usual during the launch window. One technical note worth knowing: Sonnet 5 uses an updated tokenizer, so the same text can map to roughly 1.0 to 1.35 times more tokens than before. Anthropic set the introductory pricing so that the move from 4.6 to 5 is roughly cost-neutral despite that change.
Where the numbers come from matters here. Several sites published 'Sonnet 5' specs with confident benchmark percentages and launch dates that contradicted each other for months. The figures in this guide come only from Anthropic's own launch announcement, which is the source to trust over any aggregator.
Safety: Less Sycophancy, Fewer Hallucinations
Anthropic's pre-deployment evaluations found Sonnet 5 to be an overall improvement on Sonnet 4.6 on safety. It shows lower rates of hallucination and sycophancy, meaning it makes up fewer facts and is less likely to simply agree with you when you are wrong. In agentic settings it is better at refusing malicious requests and at resisting prompt-injection attacks that try to hijack a tool-using agent. On Anthropic's broad behavioral audit it scored safer overall than 4.6, though still somewhat less aligned than the more capable Opus 4.8 and the research-only Mythos Preview.
On cybersecurity specifically, Sonnet 5 has a much lower ability than Anthropic's Opus models; in one evaluation it could never produce a full working software exploit. Even so, because it is somewhat stronger than its predecessor, Anthropic launched it with real-time cyber safeguards enabled by default, the same ones used on Opus 4.7 and 4.8, and less strict than those on Fable 5. For most users this is invisible; for anyone doing security research, Anthropic still points to Opus 4.8 for work that needs reduced guardrails.
Should You Switch, or Stay on Opus 4.8?
For most people the answer is that the switch already happened for you, and that is fine. As the new default on Free and Pro, Sonnet 5 is the right call for everyday writing, analysis, coding, and agentic tasks, and it is cheaper to run than Opus. Reach for Opus 4.8 when you need the very highest accuracy on the hardest reasoning or the most demanding autonomous coding, or for security work that needs lighter guardrails. The clean mental model: Sonnet 5 for the daily driver, Opus 4.8 for the heaviest lifting, and the effort setting to slide between them.
- On Free or Pro and just chatting or coding? You are already on Sonnet 5; no action needed, and it is a straight upgrade.
- Running agents, browsers, or terminals? Sonnet 5 is the big winner here, with much better follow-through at a lower price.
- Need maximum accuracy on the hardest tasks? Keep Opus 4.8 in reach and raise the effort level when precision matters most.
- Building on the API? Use claude-sonnet-5, and take advantage of the $2/$10 introductory pricing through August 31, 2026.
- Care about fewer made-up facts and less flattery? Sonnet 5 reports lower hallucination and sycophancy than Sonnet 4.6.
Try Sonnet 5 Against Other Models
The fastest way to judge a new model is to run your own real task through it and a couple of rivals, rather than trusting any single launch chart. A tool like LumiChats lets you reach Claude, including the latest Sonnet, alongside GPT-class and Gemini-class models and 40-plus more in one place, so you can paste the same prompt into several and see where Sonnet 5's agentic gains actually show up for the kind of work you do, before you commit your workflow to it.
01When did Claude Sonnet 5 come out?
Anthropic released Claude Sonnet 5 on June 30, 2026. It is available across all plans from launch, is the default model for Free and Pro users, and is accessible to Max, Team, and Enterprise users, in Claude Code, and on the Claude Platform via the API model name claude-sonnet-5.
02How much does Claude Sonnet 5 cost?
Standard pricing is $3 per million input tokens and $15 per million output tokens, the same as Sonnet 4.6. Through August 31, 2026 there is an introductory rate of $2 per million input and $10 per million output. An updated tokenizer means the same text can use slightly more tokens, and the intro pricing is set to keep the transition roughly cost-neutral.
03Is Claude Sonnet 5 better than Opus 4.8?
Not overall. Anthropic positions Sonnet 5's performance as close to Opus 4.8 on many tasks at a lower price, but Opus 4.8 still leads when you need the highest accuracy on the hardest reasoning and most demanding agentic work. Sonnet 5 is the value daily driver; Opus 4.8 is the precision option.
04What is new in Claude Sonnet 5 versus Sonnet 4.6?
It is the most agentic Sonnet yet, with stronger planning, tool use, coding, and autonomous multi-step work, and it is a clear improvement on reasoning and knowledge tasks. Anthropic also reports lower hallucination and sycophancy and better resistance to prompt-injection attacks than Sonnet 4.6.
05Is Claude Sonnet 5 safe for agentic and coding work?
Anthropic's evaluations found it overall safer than Sonnet 4.6 in agentic contexts, with better refusal of malicious requests and prompt-injection resistance. It has much weaker cybersecurity capabilities than Anthropic's Opus models and ships with real-time cyber safeguards enabled by default.
The takeaway: Sonnet 5 is the kind of release that matters quietly. No price hike, a real jump in agentic capability, less sycophancy, and it is already the model most Claude users are talking to. Trust Anthropic's own numbers over the hype, let it handle your daily work, and keep Opus 4.8 for the moments that truly demand the frontier.
