Claude Sonnet 5 went live on June 30, 2026, and Anthropic did not make a quiet rollout. It is now the default model for Free and Pro users on claude.ai, and it is available everywhere else too, including Claude Code, the Claude Platform, Max, Team, and Enterprise plans.
The model it replaces, Claude Sonnet 4.6, only launched in February. That is a four and a half month gap, which is fast even by Anthropic’s recent pace. The bigger question for anyone using Claude daily is not the release date. It is what actually changed between the two models, and whether switching changes how the work gets done.
TL;DR: Claude Sonnet 5 replaces Sonnet 4.6 as Anthropic’s default Sonnet model, launching at a lower introductory price of $2 per million input tokens and $10 per million output tokens, rising to standard $3 and $15 pricing on August 31, 2026. Both models share a 1M token context window. Sonnet 5 scores higher on agentic coding and knowledge work benchmarks and shows lower hallucination rates than 4.6.
What Claude Sonnet 5 actually changes
Claude Sonnet 5 is Anthropic’s newest mid tier model, sitting between Haiku and Opus in the lineup. As of today, it is the default model for anyone on a Free or Pro claude.ai plan. Developers reach it through the API using the model string claude-sonnet-5.
According to Anthropic, Sonnet 5 is built to be its most agentic Sonnet model yet.
That phrase shows up in nearly every recent Anthropic launch post, but the specifics here are concrete. Sonnet 5 can make a plan, use a browser, run a terminal, and keep working without a human nudging it forward at every step, at a level Anthropic says used to require Opus class models just a few months ago.
Rate limits across Chat, Cowork, Claude Code, and the Claude Platform have also been increased, since higher effort levels on Sonnet 5 use more tokens per task than a quick reply does.
Pricing and the context window
Sonnet 4.6 launched in February priced at $3 per million input tokens and $15 per million output tokens, the same rate as the Sonnet model before it. Sonnet 5 opens lower, at $2 per million input tokens and $10 per million output tokens through August 31, 2026, then moves to that same $3 and $15 standard rate 4.6 already charges.
Prompt caching cuts that further, with up to 90% savings on cached input, and batch processing knocks another 50% off. US only inference is also available at 1.1x standard pricing for workloads that need to stay on domestic infrastructure.
Both models carry a 1 million token context window. That part did not change between generations.
| Feature | Claude Sonnet 4.6 | Claude Sonnet 5 |
|---|---|---|
| Release date | February 17, 2026 | June 30, 2026 |
| Context window | 1M tokens | 1M tokens |
| Standard price (input / output per million tokens) | $3 / $15 | $3 / $15 |
| Introductory price (per million tokens) | — | $2 / $10 through Aug 31, 2026 |
| Agentic coding benchmark score | 58.1% | 63.2% |
| API model string | claude-sonnet-4-6 | claude-sonnet-5 |
There has been chatter online about a new tokenizer for Sonnet 5 that would burn more tokens per response than 4.6 did. That detail traces back to reporting on Claude Fable 5, not Sonnet 5. Anthropic has not announced a tokenizer change for Sonnet 5, and nothing in the official system card or pricing page mentions one. Until Anthropic confirms it directly, that claim stays unverified.
Sonnet 5 vs Sonnet 4.6 on agentic performance
The benchmark gap is where Anthropic wants the attention. On an agentic coding benchmark cited by TechCrunch, Sonnet 5 scored 63.2%, up from Sonnet 4.6’s 58.1%. Opus 4.8 still leads at 69.2%, but the gap between Sonnet 5 and Opus has narrowed enough that Anthropic is now positioning Sonnet 5 as the practical default for most agentic work, not just the budget option.
On a knowledge work benchmark, Sonnet 5 reportedly edges past Opus 4.8 outright, though Anthropic has not published the full comparison figures for that specific result.
Anthropic’s pre deployment safety testing also found Sonnet 5 hallucinates less and shows lower sycophancy than 4.6, with better resistance to prompt injection attempts during computer use tasks. It refuses clearly malicious requests more consistently too, according to the system card summary.
What changes in daily use
I ran the same multi step task through both models this week. Pull updated numbers from a spreadsheet, draft a summary email, flag anything that looks off. Sonnet 4.6 did the job but stopped halfway through to ask whether it should continue. Sonnet 5 finished the whole chain without checking in, then noted on its own that one number looked inconsistent with the rest.
That single habit, not stopping to ask permission mid task, is the thing I noticed first and kept noticing for days afterward.
That habit is a small shift, but it changes how you write prompts. I caught myself writing shorter instructions by the third day, trusting Sonnet 5 would ask if something was genuinely ambiguous instead of pausing on anything slightly open ended.
For quick one line questions, the two models feel almost identical. The difference only shows up once a task has more than two or three steps strung together, which is exactly where Anthropic says it focused the upgrade.
Frequently Asked Questions
Does Claude Sonnet 5 use the same context window as Sonnet 4.6?
Yes. Both models support a 1 million token context window, and that did not change between generations.
Is Claude Sonnet 5 cheaper than Sonnet 4.6?
For now, yes. Sonnet 5 launches at $2 per million input tokens and $10 per million output tokens through August 31, 2026, before moving to the same $3 and $15 standard rate Sonnet 4.6 already charges.
Does Sonnet 5 use a different tokenizer than Sonnet 4.6?
Anthropic has not announced a tokenizer change for Sonnet 5. Reports describing a more token hungry tokenizer actually refer to Claude Fable 5, not Sonnet 5.
Is Sonnet 5 better than Opus 4.8?
Not across the board. Opus 4.8 still leads on most benchmarks, including agentic coding at 69.2% versus Sonnet 5’s 63.2%, though Sonnet 5 narrows that gap while costing less.
What is the API model string for Claude Sonnet 5?
Developers can access it using claude-sonnet-5 through the Claude API, Claude Code, and the Claude Platform.
Who should actually switch
If you are already on a Free or Pro claude.ai plan, there is nothing to switch. Sonnet 5 is just there now. For API users still pointed at claude-sonnet-4-6, moving over makes sense for anything agentic, including multi step coding tasks, tool use chains, or workflows that used to need Opus for reliability.
For simple one shot questions, the difference is harder to feel. Sonnet 4.6 was already accurate enough for most single turn use, and Sonnet 5’s gains show up mainly when a task has several steps strung together rather than one clean request.
The pricing makes the decision easier for now. At $2 and $10 per million tokens through the end of August, Sonnet 5 is cheaper than 4.6 ever was, for a model that scores higher across the board. That window closes at the end of summer, and the math shifts slightly once standard pricing takes over.
