Opus vs Sonnet: which Claude model should you use?
For most people, Sonnet 5 (the default) is the right pick. Reach for Opus 4.8 on hard coding or reasoning tasks where accuracy matters more than speed.
Short answer: use Sonnet 5. It's the default model on Free and Pro for a reason, and for everyday writing, questions, and drafting it's not meaningfully worse than Opus. Switch to Opus 4.8 when you're deep in a hard coding problem, a long multi-step research task, or anything where a slower, more careful answer beats a fast one. Most people never need to switch at all.
This site isn't affiliated with Anthropic. Nobody's paying us to recommend one model over the other.
The one-breath verdict
| Default on Free and Pro | Sonnet 5 |
|---|---|
| Comparative latency | Sonnet 5: Fast, Opus 4.8: Moderate |
| API price per MTok (in/out) | Sonnet $3/$15, Opus $5/$25 |
| Context window | 1M tokens, both |
| Max output | 128K tokens, both |
| Best for | Sonnet: daily work. Opus: hard coding/reasoning |
Both numbers come straight from Anthropic's models overview page, fetched July 2026.
What actually changes when you switch
The two models share the same context window (1M tokens) and the same max output (128K tokens). Opus doesn't "remember more" than Sonnet. What changes is how carefully it works through a problem, and how long that takes.
Anthropic's own comparative-latency label is "Fast" for Sonnet 5 and "Moderate" for Opus 4.8. In practice that means Opus visibly takes longer per response, especially on anything non-trivial. Part of why: Opus 4.8's "effort" setting defaults to high across every surface, API, Claude Code, and claude.ai, per that same models page. It's built to slow down and think harder by default, not to match Sonnet's speed.
Cost scales the same direction. Opus runs $5 input / $25 output per million tokens against Sonnet's $3/$15 (an introductory $2/$10 through August 31, 2026), so roughly 1.7x the price at standard API rates. That's irrelevant if you're just chatting on Free or Pro, but it matters the moment you're building something on the API.
The full lineup, briefly
"Opus" and "Sonnet" aren't the only two names in play, and it's easy to lose track of where each one sits. As of August 2026, the practical lineup is four models:
- Claude Haiku 4.5: the fastest, cheapest tier. Fine for quick, low-stakes tasks.
- Claude Sonnet 5: "the best combination of speed and intelligence," per Anthropic, and the default for Free and Pro since its June 30, 2026 release.
- Claude Opus 4.8: built "for complex agentic coding and enterprise work." Selectable on Pro and Max, slower and pricier by design.
- Claude Fable 5: Anthropic's most capable widely released model, positioned for long-running agents. This is a separate, more expensive top tier ($10/$50 per MTok), not a fancier name for Opus. Most consumer chat users never touch it.
(There's also Mythos 5, an invitation-only model for defensive cybersecurity work under Anthropic's Project Glasswing. It's out of scope for everyday use.)
The honest takeaway: most people only ever need to know Sonnet is the default and Opus is the "slow down and think harder" option. Fable and Haiku sit at the edges for specific jobs, not the everyday decision.
How to switch models, and on which plans
Click the model name
It sits next to the send button in claude.ai, showing whichever model you're currently on.
Pick from the list, or click "More models"
The default list shows a few options; "More models" reveals the full set, including Opus, Sonnet, and Haiku variants.
Send your next message
The switch applies starting with Claude's next response, mid-conversation. Same menu controls effort level and thinking mode.
Availability by plan: Free and Pro both default to Sonnet 5. Opus 4.8 is selectable on Pro and fully available on Max, which lets you pick from every current model. Team and Enterprise get the same model access as Pro/Max, seat-based, though admins can restrict specific models on Enterprise.
Anthropic doesn't publish a per-model message quota for Pro (for example, "N Opus messages per week") anywhere in its own docs, just a general weekly limit that applies across all models, so treat any specific number you see elsewhere as an outside estimate, not an official figure. Full details on what each tier includes: Claude pricing explained.
Benchmarks, with a caveat
Anthropic's own Sonnet 5 vs. Opus 4.8 comparison ships as a chart image on its Sonnet 5 announcement, not as machine- readable numbers, so treat any exact percentage you see cited (including on this page) as reported by third parties rather than confirmed verbatim from Anthropic.
Directionally, Opus reportedly leads Sonnet on hard agentic and coding benchmarks like SWE-bench Pro by single digits, roughly in line with what you'd expect from a model that defaults to more careful, higher-effort reasoning. For anything you need to cite precisely, link the model announcement and its linked system card rather than a third-party aggregator.
Tested against platform.claude.com/docs, July 2026.
If you're new to Claude generally
If you haven't settled into a workflow yet, the model question is a second-order decision. Start with how to use Claude to get the basics down, and check Claude pricing before you decide whether Pro or Max is worth it. If you're weighing Claude against ChatGPT more broadly, our comparison covers that separately from the model-picking question here.
Is Opus better than Sonnet?
Opus 4.8 reasons more carefully and reportedly leads on hard coding and agentic benchmarks, but it's slower ("Moderate" latency vs. Sonnet's "Fast") and about 1.7x the API price. "Better" depends on the task: for daily writing and quick questions, most people can't tell the difference in the output that matters to them.
Which Claude model is the default?
Sonnet 5, on both the Free and Pro plans, since its June 30, 2026 release. Max and Team/Enterprise let you pick from the full model lineup, including Opus, without a fixed default.
Do I need Opus?
Probably not for daily use. Reach for it specifically on long, hard coding tasks or multi-step reasoning where a slower, more thorough answer is worth the wait. If you're not sure, start on Sonnet (the default) and switch only when you hit a task that feels like it needs more care.
Is Opus more expensive than Sonnet?
Yes, on the API: $5/$25 per million input/output tokens for Opus 4.8 versus $3/$15 for Sonnet 5 ($2/$10 introductory through August 31, 2026), about 1.7x the cost. On Free, Pro, or Max chat plans, there's no separate per-model charge, just your existing subscription and its shared usage limits.