Claude Opus 5.5: What's New, Pricing & How to Use It for Vibe Coding
Claude Opus 5.5 brings Fable-level coding at a lower price. Here's what's new, the benchmarks, API pricing, effort levels, how to use it in Claude Code, and how it compares to GPT-6 and Gemini 4.
On this page (6)
Claude Opus 5.5 is Anthropic's newest Opus model, released on September 22, 2026. Anthropic says it performs at the level of its pricier Claude Fable 5.1 on most work, runs more than 30% faster than Opus 5, and costs 20% less per token. It's now the default model in Claude Code, which makes it the model many vibe coders are using right now, whether they chose it or not.
This guide covers what changed, what it costs, and how to get the best results from it. For the wider picture of this month's launches, see our October 2026 AI news roundup.
Opus 5.5 at a glance
| Claude Opus 5.5 | |
|---|---|
| Released | September 22, 2026 |
| API model ID | claude-opus-5-5 |
| Input price | $4 per million tokens (was $5 on Opus 5) |
| Output price | $20 per million tokens (was $25) |
| Cache reads | $0.20 per million tokens (60% cheaper) |
| Max output | 128,000 tokens |
| Fast mode | Up to ~2.5x faster at $8 / $40 per million tokens |
| Available on | Claude apps (paid plans), Claude Code, Claude API, AWS, Google Cloud, Microsoft Azure |
Source: Anthropic's Opus 5.5 announcement (opens in a new tab). Anthropic estimates typical workloads cost about 40% less than on Opus 5 once you factor in cheaper caching and better token efficiency.
What's new
Better at agentic coding
On Anthropic's published benchmarks, Opus 5.5 beats both Opus 5 and Fable 5.1 on agentic coding:
| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 66.4% | 55.8% | 52.3% |
| FrontierCode v1.1 | 54.4% | 50.3% | 48.0% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% |
Benchmarks published by a model's creator should be read with healthy skepticism, but early hands-on reports broadly match: it's strong at long, multi-step tasks like migrations and refactors. Anthropic highlights a customer migrating 680,000 lines of code in under a day.
Faster and more concise
Output is over 30% faster than Opus 5, and Anthropic says the model "puts the most important information up front" with less padding. In practice that means shorter, more scannable summaries at the end of agent tasks.
Effort levels
Opus 5.5 uses adaptive thinking with selectable effort: low, medium, high, xhigh and max. Higher effort means more reasoning, more tokens and more cost. Simon Willison noted (opens in a new tab) that at max effort it can over-think simple prompts until it hits its output limit, so max isn't automatically better.
Safety
Anthropic calls Opus 5.5 the best-aligned model it has tested, reports it attempts to circumvent sandbox boundaries 85% less often than Opus 5, and says prompt-injection resistance matches or beats Opus 5. Because its biology and cybersecurity capabilities are high, some requests in those areas are restricted or routed differently unless you're enrolled in Anthropic's verification programs.
How to use Opus 5.5 for vibe coding
In Claude Code
Opus 5.5 is the default for most paid plans. To check or switch, run /model inside a session. Update first if you don't see it:
claude update
Pick the right effort level
- Low/medium: quick edits, small bug fixes, explaining code.
- High (a good default): features, multi-file changes, debugging.
- xhigh/max: genuinely hard problems like tricky concurrency bugs or large migrations, and only after a planning step.
If you're on a usage-limited plan, lower effort stretches your allowance noticeably.
Use plan mode for big tasks
Opus 5.5 is excellent at long-horizon work, but only when it's pointed in the right direction. Start large tasks in plan mode (Shift+Tab in Claude Code), review the plan, then let it run. Our spec-driven development guide explains the workflow.
Give it context
A concise AGENTS.md or CLAUDE.md still beats any model upgrade for keeping an agent on track in your codebase.
Watch out for over-thinking
If a simple task is taking minutes, lower the effort or break the request down. Clear, specific prompts help too; see our prompting playbook.
Opus 5.5 vs the competition
| Model | Input / output per 1M tokens | Positioning |
|---|---|---|
| Claude Opus 5.5 | $4 / $20 | Frontier coding at a mid-range price |
| Claude Fable 5.1 | $10 / $50 | Anthropic's top tier for the longest, hardest work |
| GPT-6 Astra | $10 / $50 | OpenAI flagship, powers Dots |
| GPT-6 Sol | $2 / $10 | OpenAI's everyday workhorse, Codex default |
| Gemini 4 Argon | $2 / $10 intro, then $4 / $20 | Google's new frontier model |
Our full comparison, best AI model for coding in 2026, goes deeper.
Should you switch?
- Already on Claude Code? You probably are using it already. Just update.
- Paying per token via the API? Moving from Opus 5 is an easy cost saving. Moving from Fable 5.1 is worth testing on your own tasks; many will see similar quality for far less.
- Using Cursor or Copilot? Opus 5.5 is available in their model pickers; try it on your next multi-file feature.
- On a free plan? Opus 5.5 isn't available on Claude's free tier. Look at free AI coding options instead.
Frequently asked questions
When was Claude Opus 5.5 released?
September 22, 2026.
How much does Claude Opus 5.5 cost?
$4 per million input tokens and $20 per million output tokens via the API. In the Claude apps and Claude Code it's included in paid plans like Pro and Max, subject to usage limits.
Is Opus 5.5 better than Fable 5.1?
Anthropic says it matches Fable 5.1 on most work and beats it on several agentic coding benchmarks, at well under half the price. Fable 5.1 remains Anthropic's top tier for the most demanding long-running tasks.
Are Sonnet 5.5 and Haiku 5.5 coming?
Anthropic has indicated that Sonnet 5.5 and Haiku 5.5 will follow within weeks of Opus 5.5. We'll cover them when they ship.
- #Claude Opus 5.5
- #Anthropic
- #Claude Code
- #AI Models
- #AI News