Updated 30 September 2026: Anthropic released Claude Sonnet 5.5 on 28 September at the same $2 / $10 price as Sonnet 5 but with Opus 5.5’s June 2026 knowledge cutoff, so we have rewritten this comparison around it.
Of all the Claude models Anthropic sells, most people only ever choose between two of them. Sonnet is the fast one and Opus is the powerful one, and since 28 September 2026 that means Claude Sonnet 5.5 and Claude Opus 5.5. The gap between them is now small enough that the usual advice, “Opus if you can afford it”, no longer obviously holds. So if you’re weighing up Claude Sonnet vs Opus, here’s what actually separates them, what each costs in the API and in the Claude apps, and when the cheaper model is enough for the work you do.
- Sonnet 5.5 is the default for most people. It has the same 1M-token context, 128K output and June 2026 knowledge cutoff as Opus, costs half as much and runs faster. Anthropic calls it “a faster, lower-cost complement to Claude Opus 5.5”.
- Opus 5.5 earns its price on long, sprawling jobs. Think multi-hour coding agents, codebase-wide migrations, or open-ended work where the model has to keep making judgement calls. At $4 in and $20 out per million tokens it’s 20 percent cheaper per token than the Opus 5 it replaces, and Anthropic says it costs 40 percent less on typical workloads.
- In the Claude apps the question mostly goes away. The Free plan gives you Sonnet 5.5. Pro, at $20 a month, adds Opus, and you can switch models per conversation.
- Try the effort dial before you switch models. Both models have an effort setting, and Anthropic’s own guidance says tuning it “is often a better lever than switching models”.
The current models and prices
Claude Opus 5.5 came out on 22 September 2026 and took over from Opus 5 as the current Opus. Claude Sonnet 5.5 followed six days later, on 28 September, and replaced Sonnet 5, which had only been out since 30 June. Opus 5 and Sonnet 5 are both still served, but Anthropic now lists them as legacy models. Above both sits Claude Fable 5.1, Anthropic’s most capable model at $10 and $50 per million tokens. Anthropic pitches it at “demanding reasoning and long-horizon agentic work”, and we’ve left it out of this comparison. Claude Haiku 5.5 completes the 5.5 line-up as the cheap option.
| Model | Price per 1M tokens (in / out) | Context / max output | Latency | Default effort | Knowledge cutoff |
|---|---|---|---|---|---|
 Claude Sonnet 5.5claude-sonnet-5-5 · 28 Sep 2026 | $2 / $10 | 1M / 128K | Fast | high | Jun 2026 |
| Claude Opus 5.5claude-opus-5-5 · 22 Sep 2026 | $4 / $20 | 1M / 128K | Moderate | medium | Jun 2026 |
| Claude Sonnet 5legacy | $2 / $10 | 1M / 128K | Fast | high | Jan 2026 |
| Claude Opus 5legacy | $5 / $25 | 1M / 128K | Moderate | high | May 2026 |
| Claude Haiku 5.5claude-haiku-5-5 · 7 Oct 2026 | from $0.10 / $0.50 | 1M / 128K | Fastest | medium | Jun 2026 |
Two details in that table matter more than the others. Sonnet 5.5 costs half what Opus 5.5 does on both input and output, at the same $2 and $10 that Sonnet 5 charged. Anthropic first announced that rate as introductory for Sonnet 5 and has since confirmed it as the standard price. The other detail is the knowledge cutoff. Sonnet 5 stopped at January 2026, which was one of the clearest reasons to pay for Opus. Sonnet 5.5 and Opus 5.5 both run to June 2026, so on recent events the two now know the same things. Haiku 5.5, released on 7 October 2026, is the budget pick for high-volume work. Its price is tiered: $0.10 in and $0.50 out per million tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 above that.
What Opus is better at
Anthropic is unusually specific about this. It describes Opus 5.5 as “for long-running agentic coding and knowledge work”, meaning jobs where a model works for an hour or more, calling tools, reading files and revising its own plan as it goes. In the launch material, “long and sprawling jobs like codebase-wide migrations and audits” come first on the list. The Sonnet 5.5 announcement draws the same line from the other side, saying “Opus 5.5 remains clearly stronger at complex, open-ended work requiring sustained judgment”. Efficiency is the other part worth dwelling on. GitHub, one of Anthropic’s early testers, says that in its testing Opus 5.5 “used among the fewest tokens and steps we measured”. A model that finishes a job in fewer steps can work out cheaper per task than one with a lower price per token. Opus 5.5 is also the default model in Claude Code, which we compare with Cursor in Cursor vs Claude Code.
This time Anthropic has put the two side by side itself. In the Sonnet 5.5 launch table, Sonnet 5.5 scores 1844 on GDPval-AA, a test of real-world office work, against 1846 for Opus 5.5. It gets 55.5 percent on CursorBench 4.0 against 57.8, and 80.1 percent on OSWorld 2.1 against 81.8. On Terminal-Bench 4.0 it actually comes out ahead, at 70.6 percent against Opus’s 66.4. Anthropic reads its own numbers more cautiously than that, though. It says “benchmark scores capture only one facet of a model’s capabilities”, and it pitches Sonnet 5.5 at “well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets”, leaving the open-ended work to Opus. It’s also a big step up from the Sonnet it replaces. On Terminal-Bench 4.0, Sonnet 5 scored 10.3 percent.
Where they’re the same
In more places than you’d expect. Both take 1 million tokens of context, which is roughly 555,000 words, and both can write up to 128,000 tokens in a single reply. Both have a reliable knowledge cutoff of June 2026. Both accept text and images. Both use adaptive thinking, where the model decides how much to reason before it answers, guided by the effort setting. And both run on every platform Anthropic supports, meaning its own API, Amazon Bedrock, Google Cloud and Microsoft Foundry. Cache reads cost $0.20 per million on either. If you ask one question and want one answer, you often won’t be able to tell them apart. The difference shows up on long, open-ended jobs.
+Pick Sonnet 5.5 when
- You pay per token, since it costs half as much as Opus 5.5
- Speed matters, as in chat, support or anything a person is waiting on
- The task has clear limits, such as drafting, summarising, extraction, bug fixes or everyday coding
- You run high volume and can batch it, which brings the price to $1 / $5 with the 50 percent batch discount
+Pick Opus 5.5 when
- The job runs for hours, as with migrations, audits or autonomous coding agents
- The work is open-ended and needs sustained judgement, where Anthropic says Opus stays “clearly stronger”
- Getting a task done in fewer steps matters more than the price per token
- Sonnet’s answers are good but not quite right on your evals
How the effort setting fits in
Both models have an effort parameter with five levels, from low to max, which trades thinking depth against speed and cost without leaving the model. On the API, Sonnet 5.5 defaults to high. Opus 5.5 defaults to medium, one notch lower, and that’s part of why it’s both cheaper and faster than the model it replaced. In Claude Code and the Claude apps, Anthropic runs Sonnet 5.5 at medium too. If you tuned effort on Sonnet 5, start again, because Anthropic says Sonnet 5.5’s levels “are recalibrated” and the same setting no longer produces the same amount of thinking. Anthropic puts it plainly in its own model-selection guide: “Tuning effort is often a better lever than switching models.” So the fairer comparison is Sonnet at high effort against Opus at medium, and for a lot of work those two end up closer than the price list suggests.
Prices in the Claude apps
If you use Claude through the website or the apps rather than the API, picking a model mostly comes down to picking a plan. The Free plan runs Sonnet and Haiku, and Anthropic says “anyone can chat with Claude using Sonnet 5.5”, so free users get the new model too. Pro costs $20 a month, or $17 on an annual plan, and adds Opus plus “at least 5x more usage per 5-hour session than Free”. Max starts at $100 a month, gives you 5x or 20x the usage of Pro and includes partial access to Fable. On a paid plan you switch models from the menu at the top of a conversation. Choosing Opus costs nothing extra, though it does get through your allowance faster. If you’re still choosing an assistant, we compare Claude Pro with ChatGPT Plus and Google AI Pro in Claude, ChatGPT or Gemini.
Working out the API bill
Say you run a support assistant that handles 10,000 conversations a month, each about 3,000 input and 700 output tokens. On Sonnet 5.5 that’s 30 million input tokens at $2 and 7 million output tokens at $10, so $60 plus $70, or $130 a month. The same traffic on Opus 5.5 is 30 million at $4 and 7 million at $20, so $120 plus $140, or $260. Batch processing halves both figures, to $65 and $130, for anything that doesn’t need an instant reply, and prompt caching trims the input side further on both models at the same $0.20 cache-read rate. Those are list prices per token. Anthropic also says Sonnet 5.5 “typically needs far fewer tokens to do the same work” and costs “up to 30% less per task” than Sonnet 5, so if you’re moving across from Sonnet 5 the real bill may come in lower. For a workload like this we’d pick Sonnet without much thought. Opus is a $130-a-month experiment you could run for a week to see whether the answers get better.
A coding agent that refactors a large codebase overnight is a different story. What drives that bill is how many steps, tool calls and re-reads the agent needs to finish, and the per-token price matters much less. GitHub’s finding that Opus 5.5 “used among the fewest tokens and steps” it measured is exactly the kind of evidence that counts here. Sonnet 5.5 makes a similar pitch, with Anthropic saying early testers saw it batch tool calls “leading to fewer steps and lower costs”. Neither claim settles the question for your code, but you can test both models on your own repository in an afternoon.
Frequently asked questions
Is Claude Opus better than Sonnet?
Yes, but by less than ever. Anthropic says Opus 5.5 “remains clearly stronger” on complex, open-ended work, while Sonnet 5.5 comes within a few points of it on most of Anthropic’s published benchmarks and both know events up to June 2026. For single questions, drafting and everyday coding, Sonnet 5.5 is close enough that most people won’t notice, and it costs half as much.
How much more does Opus cost than Sonnet?
Exactly double on the API. Opus 5.5 is $4 per million input tokens and $20 per million output tokens, while Sonnet 5.5 is $2 and $10, the same as Sonnet 5. Cache reads cost $0.20 on both. In the Claude apps, paid plans don’t charge extra for Opus, but it uses up your allowance faster.
Can I use Opus for free?
Not on claude.ai. The Free plan runs Sonnet 5.5 and Haiku, and Opus is available from the Pro plan at $20 a month. New API accounts do get a small amount of free credit, which works with any model.
Does Sonnet have a smaller context window?
No. Sonnet 5.5 and Opus 5.5 both take 1 million tokens of context and return up to 128,000 tokens per reply. Haiku 5.5 has the same limits too. Only the older Haiku 4.5 is smaller, at 200,000 tokens of context.
What about Claude Fable?
Claude Fable 5.1 is the model above Opus, at $10 and $50 per million tokens. Anthropic only recommends it when Opus 5.5 at its highest effort levels still falls short on demanding reasoning or long agent runs. Pro plans can use it through usage credits, and Max plans can spend up to half the weekly limit on it.
Sources and method
We took prices, context windows, effort defaults, knowledge cutoffs and release dates from Anthropic’s developer documentation and consumer pricing page on 28 September 2026, and checked all of them again on 30 September 2026 after Sonnet 5.5 came out. We re-checked the current Claude line-up on 28 October 2026. Benchmark figures and quotations come from Anthropic’s own launch announcements and documentation. They are the company’s claims, and we haven’t run an independent head-to-head test for this guide. The cost examples use list prices only.
- Anthropic, Models overview
- Anthropic, Pricing
- Anthropic, Choosing the right model
- Anthropic, Claude Sonnet 5 model page
- Anthropic, Claude Opus 5.5 model page
- Anthropic, Introducing Claude Sonnet 5
- Anthropic, Introducing Claude Opus 5.5
- Anthropic, Introducing Claude Sonnet 5.5 (checked 30 September 2026)
- Anthropic, Claude Sonnet 5.5 model page (checked 30 September 2026)
- Anthropic, What’s new in Claude Sonnet 5.5 (checked 30 September 2026)
- Anthropic, Effort (checked 30 September 2026)
- Anthropic, Claude Sonnet product page (checked 30 September 2026)
- Claude, consumer plans and pricing
