# Sonnet 5.5 vs Opus 5.5: which Claude model should you use?

Half the price, close on most coding benchmarks. Where Opus still earns its cost, from Anthropic's launch posts and docs.

Published 2026-10-05, updated 2026-10-07 by Richard Kaminsky and Mitchell Lipyansky, the co-founders of Poly. Canonical: https://usepoly.co/sonnet-5-5-vs-opus-5-5
Poly is a multiplayer AI coding workspace: a shared room where your team works with one AI agent, together. Free to start: https://usepoly.co/

**Claude Sonnet 5.5 costs half as much as Opus 5.5 per token, $2 per million input tokens and $10 per million output against $4 and $20, and on Anthropic's own coding benchmarks it comes close: 55.5% to 57.8% on CursorBench 4.0, and 70.6% to 66.4% on Terminal-Bench 4.0, where Sonnet came out ahead. Anthropic still says Opus 5.5 is "clearly stronger at complex, open-ended work requiring sustained judgment," and its docs suggest starting with Opus when unsure. Both have a 1 million token context window and 128,000 output tokens. Use Sonnet 5.5 for everyday coding and fast agent work, and Opus 5.5 for long, ambiguous tasks where a wrong turn costs more than the tokens.**

## Key takeaways

- Sonnet 5.5 is $2 / $10 per million tokens and Opus 5.5 is $4 / $20; both have 1M context and 128K output. Opus 5.5 is 20% cheaper per token than Opus 5 was.
- On Anthropic's benchmarks they are close; on Vals AI's independent Terminal-Bench 4.0 run they are level, at 64.1% and 65.2%.
- Claude Code defaults to Opus 5.5 at medium effort on paid plans and the API. Anthropic says Opus 5.5 at medium matches or beats Opus 5 at high.
- Claude's free plan includes Sonnet, not Opus.

*Claude Sonnet 5.5 and Opus 5.5, from Anthropic's launch posts and docs, October 2026.*

|  | Sonnet 5.5 | Opus 5.5 |
| --- | --- | --- |
| Released | September 28, 2026 | September 22, 2026 |
| Price per million tokens (input / output) | $2 / $10 | $4 / $20 |
| Context window / max output | 1M / 128K | 1M / 128K |
| Default effort (API / Claude Code) | High / medium | Medium / medium |
| Terminal-Bench 4.0 | 70.6% | 66.4% (xhigh) |
| CursorBench 4.0 | 55.5% | 57.8% |
| FrontierCode 1.1 | 46.2% (52.1% at xhigh) | 54.4% |
| OSWorld 2.1, partial | 80.1% | 81.8% |
| On Claude's free plan | Yes | No |
| Anthropic's description | Everyday coding, agent and enterprise work | Complex agentic coding and enterprise work |

Sources: Anthropic's [Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) and [Opus 5.5](https://www.anthropic.com/news/claude-opus-5-5) launch posts and [API pricing](https://platform.claude.com/docs/en/about-claude/pricing). Benchmarks at max effort unless noted.

## How much do Sonnet 5.5 and Opus 5.5 cost?

Per million tokens on Anthropic's API, Sonnet 5.5 is $2 for input and $10 for output, and Opus 5.5 is $4 and $20. Cached input costs $0.20 on Opus 5.5 and $0.10 on Sonnet 5.5, which Anthropic cut from $0.20 on October 7, 2026 ([Anthropic](https://platform.claude.com/docs/en/about-claude/pricing)). Batch processing halves the price, and the full 1M-token window carries no surcharge. Opus 5.5 also has a fast mode at $8 / $40 for up to 2.5 times the speed.

Price per token isn't cost per task, because models use different numbers of tokens. Anthropic says Sonnet 5.5 costs up to 30% less per task than Sonnet 5, and Opus 5.5 about 40% less than Opus 5 on typical workloads. On [Vals AI's Terminal-Bench 4.0](https://www.vals.ai/benchmarks/terminal-bench-4) run, an average Sonnet 5.5 task cost $16.51 and an Opus 5.5 task $13.20, so the cheaper model per token was the dearer one per task there.

## How do they compare on benchmarks?

From Anthropic's launch posts, Sonnet 5.5 against Opus 5.5:

- **Terminal-Bench 4.0:** 70.6% against 66.4%, Opus measured at xhigh effort. Anthropic gives a standard error of about 2.6 points.
- **CursorBench 4.0:** 55.5% against 57.8%.
- **FrontierCode 1.1:** 46.2% at max effort, or 52.1% at xhigh, against 54.4%. Anthropic notes Sonnet scores lower at max than at xhigh.
- **OSWorld 2.1 (partial):** 80.1% against 81.8%.
- **GDPval-AA v2.1:** 1844 against 1846 Elo.

Anthropic's 2026 posts don't report SWE-bench, so treat SWE-bench figures for these models elsewhere as third-party. The one independent run we could check, Vals AI's Terminal-Bench 4.0 (updated October 1, 2026), put Opus 5.5 at 65.15% and Sonnet 5.5 at 64.14%: level, not a Sonnet lead.

## When should you use Sonnet 5.5, and when Opus 5.5?

Anthropic's [model guide](https://platform.claude.com/docs/en/about-claude/models/choosing-a-model) describes Sonnet 5.5 as speed and capability for everyday coding, agent and enterprise work, and Opus 5.5 as the model for complex agentic coding, with Opus as the place to start when you're unsure. In practice:

- **Sonnet 5.5:** everyday coding with a clear scope, quick back-and-forth, high-volume agents and anything where speed matters. It's about 30% faster than Sonnet 5.
- **Opus 5.5:** long-running agent work, ambiguous problems and changes to a large codebase where a mistake is expensive.
- **Fable 5.1** ($10 / $50): Anthropic's most capable model, for demanding reasoning and long-horizon work when Opus at higher effort still falls short.
- **Haiku 5.5** ($0.10 / $0.50, five times that for prompts over 100K tokens): the fastest and cheapest, with a 1M window and, for the first time on a Haiku, effort levels. Released October 7, 2026, it replaced Haiku 4.5 ($1 / $5) ([Anthropic](https://platform.claude.com/docs/en/about-claude/pricing)).

## What do effort levels do?

Effort sets how many tokens Claude spends, thinking included, and lower effort also means fewer, terser tool calls ([Anthropic](https://platform.claude.com/docs/en/build-with-claude/effort)). For Opus 5.5, Claude Code's [docs](https://code.claude.com/docs/en/model-config) recommend:

- **Medium,** the default on Opus 5.5, for day-to-day engineering with a clear scope. Anthropic says Opus 5.5 at medium matches or exceeds Opus 5 at high on coding and knowledge-work evaluations.
- **High** where verification matters or edge cases are likely, such as fixing a bug in an existing codebase.
- **Max** rarely: the docs say it may show diminishing returns and is prone to overthinking.

Opus 5.5's thinking can't be switched off. For Sonnet 5.5, Anthropic suggests starting agent work at medium and moving to high for harder tasks.

## Which one does Claude Code use?

Opus 5.5, by default, on Pro, Max, Team, Enterprise and the API. The aliases `opus` and `sonnet` point to Opus 5.5 and Sonnet 5.5, `/model` switches mid-session, and `/effort` changes the effort level. The `opusplan` setting uses Opus in plan mode and Sonnet to carry out the plan.

## Which Claude plans include each?

Claude's free plan includes Sonnet and Haiku but not Opus. Pro, Max, Team and Enterprise include both. Fable 5.1 costs usage credits on Pro and on standard Team seats, while Max plans and premium seats can spend up to half their weekly limits on it ([Anthropic](https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan)). Anthropic publishes no fixed ratio for how much faster Opus uses a plan's limits than Sonnet; it says usage depends on the model and the effort level.

## Can you use both in one place?

In Claude's own apps you switch with the model picker. In Poly (usepoly.co), Sonnet 5.5 is on the free tier and Opus 5.5 and Fable 5.1 on paid plans, and each person in a room picks their own model and effort for their own turns in the same conversation, alongside GPT, Grok, Kimi and Muse. What a switch costs is covered in [What happens when you switch AI models mid-conversation?](/switch-ai-models-mid-conversation), and the wider comparison in [Claude vs ChatGPT for coding](/claude-vs-chatgpt-for-coding).

## Common questions

**Is Sonnet 5.5 better than Opus 5.5?**

Not overall. On Anthropic's benchmarks Sonnet 5.5 leads on Terminal-Bench 4.0 and trails on CursorBench, FrontierCode and OSWorld, and Anthropic calls Opus 5.5 clearly stronger at complex, open-ended work. Sonnet costs half as much per token.

**How much cheaper is Sonnet 5.5 than Opus 5.5?**

Half the price per token: $2 and $10 per million input and output tokens, against $4 and $20. Cost per task depends on how many tokens each uses; on Vals AI's Terminal-Bench run, Sonnet's tasks cost more on average.

**Is Opus 5.5 free?**

Not on Claude's free plan, which includes Sonnet and Haiku. Opus 5.5 comes with Pro and higher plans. In Poly, Sonnet 5.5 is on the free tier and Opus 5.5 on paid plans.

**What is the difference between Opus 5.5 at medium and high effort?**

Medium, the default, spends fewer tokens and suits clearly scoped work; Anthropic says it matches or beats Opus 5 at high. High spends as many tokens as the task needs and suits work where verification matters, such as bugs in an existing codebase.

More guides: https://usepoly.co/guides · Security: https://usepoly.co/security · Pricing: https://usepoly.co/pricing