The Claude 5.5 family: Opus, Sonnet or Haiku for coding?
Read in : Français

Contents
- What shipped, and when?
- How much do Opus, Sonnet and Haiku 5.5 cost?
- What do the coding scores say?
- What vibe coding platforms are saying
- Which Claude 5.5 model should you use for which task?
- How do you switch models in Claude Code?
- What to do now
- Frequently asked questions
- Frequently asked questions
- Sources
The Claude 5.5 family has been complete since October 7, 2026: Anthropic shipped Opus 5.5, then Sonnet 5.5, then Haiku 5.5 in just over two weeks. For vibe coders the question is no longer "which model is best" but "which model for which job," because input prices across the three span a 40x range. Here are the official numbers and a simple rule for choosing. If terms like tokens and agents are new to you, start with the complete vibe coding guide.
What shipped, and when?
Anthropic released three models in sixteen days, each with its own announcement page:
- Claude Opus 5.5, September 22, 2026, billed as the first model in the 5.5 family. Anthropic says it "performs at the level of Claude Fable 5.1 on most work" and costs 40% less to run than Opus 5.
- Claude Sonnet 5.5, September 28, 2026. Anthropic writes that its "jump in performance is particularly noticeable in coding."
- Claude Haiku 5.5, October 7, 2026, about 75% cheaper to run on average than Haiku 4.5, according to Anthropic.
On the same day, Anthropic halved Sonnet 5.5 cache-read pricing and added monthly API credits to Max and Team subscriptions: $100 or $200 depending on the individual tier, and up to $500 pooled for Team. Simon Willison, who analyzed the announcement, points out that these credits "do not roll over": whatever you don't use in a month is gone.
How much do Opus, Sonnet and Haiku 5.5 cost?
All three are priced per million tokens, with a 40x gap between Haiku and Opus on input. Prices taken from Anthropic's official pages.
| Model | Input | Output | Cache read | API ID |
|---|---|---|---|---|
| Opus 5.5 | $4 | $20 | $0.20 | claude-opus-5-5 |
| Sonnet 5.5 | $2 | $10 | $0.10 | claude-sonnet-5-5 |
| Haiku 5.5 (up to 100,000 tokens) | $0.10 | $0.50 | $0.01 | claude-haiku-5-5 |
| Haiku 5.5 (over 100,000 tokens) | $0.50 | $2.50 | $0.05 | claude-haiku-5-5 |
Two traps to know about. Haiku 5.5 gets 5x more expensive once your prompt passes 100,000 tokens. And Haiku 5.5 uses a new tokenizer: Simon Willison measured that the same long prompt produces about 1.25x as many tokens as with Haiku 4.5, which he calls "a hidden price increase."
For context, GPT-6.1 Sol, launched by OpenAI on September 29, costs the same as Sonnet 5.5 ($2 / $10).
What do the coding scores say?
On Terminal-Bench 4.0, Sonnet 5.5 scores 70.6%, ahead of Opus 5.5 at 66.4% and Haiku 5.5 at 39.2%, according to Anthropic's published tables. On FrontierCode 1.1 and CursorBench 4.0, Opus 5.5 stays narrowly ahead.
| Benchmark (Anthropic's numbers) | Opus 5.5 | Sonnet 5.5 | Haiku 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 66.4% | 70.6% | 39.2% |
| FrontierCode 1.1 (Main) | 54.4% | 52.1% | 46.4% |
| CursorBench 4.0 | 57.8% | 55.5% | not published |
These are vendor scores, often at maximum or "xhigh" effort. Anthropic itself notes, for Opus 5.5, a standard error of about 2.6 points on Terminal-Bench. Take the order of magnitude, not the decimal.
What vibe coding platforms are saying
App-building platforms shared early feedback on Anthropic's pages. These quotes were picked by the vendor, so they lean positive, but they come with numbers.
Across 118 real app builds, Claude Sonnet 5.5 produced apps that scored level with Opus 5.
— Gabriel Grinberg, Base44, on Anthropic's Sonnet 5.5 page
For Lovable builders, Opus 5.5 means faster builds with the same quality.
— Fabian Hedin, Lovable, on Anthropic's Opus 5.5 page
Lovable's CTO adds that Opus 5.5 finishes in a third to half fewer steps. If you use Lovable or Base44, you get these models without changing anything: the platform picks. The choice becomes yours in Claude Code, Cursor or the API.
Which Claude 5.5 model should you use for which task?
The simplest rule: Sonnet 5.5 by default, Opus 5.5 for big jobs, Haiku 5.5 for small repetitive work. Here is the decision tree.
- If you are adding a feature, fixing a bug or reworking a screen, then use Sonnet 5.5: top Terminal-Bench score at half the price of Opus.
- If you are running a migration, a full-codebase audit or a multi-hour task, then use Opus 5.5: Anthropic describes it as "particularly good at long and sprawling jobs like codebase-wide migrations and audits."
- If you need files summarized, bulk renames or subagents on narrow tasks, then use Haiku 5.5: Anthropic recommends it for "compaction, summarization, and subagent work" and says Sonnet and Opus remain better for complex agentic coding.
- If your app calls the API itself (a chatbot, message triage), then start with Haiku 5.5 and move up only if quality falls short.
To see how an agent splits work across models, read our explainer on coding agents and subagents.
How do you switch models in Claude Code?
Claude Code uses the /model command during a session, or the --model flag at launch, according to its official documentation. The opus, sonnet and haiku aliases point to the latest version on the Anthropic API.
/model sonnet
/model opus
/model opusplan
/effort medium
The opusplan alias uses Opus in plan mode, then switches to Sonnet to write the code: often the best trade-off when you plan a feature before having it built. The /effort command sets how hard the model thinks; Anthropic says Sonnet 5.5 defaults to "medium" effort in Claude Code. If you're new to it, follow our beginner's guide to Claude Code.
What to do now
Run a test on your own project rather than trusting leaderboards.
- Pick a real, bounded task (a known bug, a small screen to add).
- Create one Git branch per model so you can compare and roll back.
- Run the same instruction with Sonnet 5.5, then Opus 5.5.
- Note the time, the cost shown by
/costand how many fixes you needed. - Keep Sonnet as your default unless Opus is clearly better on your kind of task.
To budget the month, use our breakdown of what a vibe-coded app really costs; to see how Claude stacks up against Codex, read our coding agents comparison.
Frequently asked questions
Frequently asked questions
What is the difference between Claude Opus 5.5 and Sonnet 5.5?
Opus 5.5 costs twice as much as Sonnet 5.5 ($4 versus $2 per million input tokens) and is aimed at long, sprawling tasks such as codebase migrations. Yet Sonnet 5.5 scores higher on Terminal-Bench 4.0 (70.6% versus 66.4%, according to Anthropic). For everyday vibe coding, Sonnet 5.5 offers the best value.
Is Claude Haiku 5.5 good enough to build an app?
Not for building a whole app. Haiku 5.5 scores 39.2% on Terminal-Bench 4.0, versus 70.6% for Sonnet 5.5, and Anthropic recommends it for summarization, context compaction and subagents. It is, however, a strong choice as the model your app calls, for example to sort or summarize messages, at $0.10 per million input tokens.
Why can Haiku 5.5 cost more than expected?
Two reasons. Haiku 5.5 pricing jumps 5x once the prompt passes 100,000 tokens. And its new tokenizer produces about 1.25x as many tokens as Haiku 4.5 for the same text, according to Simon Willison's measurements. Watch your prompt sizes and the actual token counter.
How do I use Sonnet 5.5 in Claude Code?
Type /model sonnet in a Claude Code session, or launch with claude --model sonnet. The sonnet alias points to the latest Sonnet on the Anthropic API. You can also use opusplan, which plans with Opus and codes with Sonnet. The /effort command then sets how deeply the model reasons.


