Claude Just Had Its Biggest Month Yet: Opus 5.5, Haiku 5.5, Duet, and the Best Way to Learn It All

If you blinked in September, you missed a new Claude model. Between September 22 and October 7, Anthropic shipped three of them (Opus 5.5, Sonnet 5.5, and Haiku 5.5), put Cowork into every conversation, made Design, Slides, and Docs free for everyone, and then capped it off with Claude Duet, an app where Claude works ahead of you instead of waiting for a prompt.

That's a lot. And it's genuinely exciting, because this round is less about benchmark bragging and more about Claude getting cheaper, faster, and more useful in the places you already work.

This article is the practitioner's cut: what shipped, what's worth your attention, a workflow you can steal today, and the courses (free and paid) that will get you fluent fastest.

What shipped: the fall 2026 timeline

Every date below comes from Anthropic's own release notes.

The models: Opus 5.5, Sonnet 5.5, Haiku 5.5

Opus 5.5: frontier smarts, without frontier pricing

Opus 5.5 is the headline. Anthropic says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. On the API it's $4 per million input tokens and $20 per million output tokens, with a 1M-token context window and up to 128k tokens of output.

It's now the default in the Claude app, Claude Code, and Cowork on Pro, Max, and Team plans. In practice that means the model doing your code reviews, research synthesis, and long documents just got a big upgrade, and you didn't have to change a setting.

Sonnet 5.5: the everyday default for builders

Sonnet 5.5 is the second model in the family and the one most production apps will reach for: faster and cheaper than Opus 5.5, still very capable. Anthropic also cut Sonnet 5.5's cache-read price in half on October 7, which matters a lot if your app reuses a big system prompt or document set.

Haiku 5.5: cheap, fast, and surprisingly capable

Haiku 5.5 is the one to watch if you run anything at volume: classification, extraction, routing, triage, summaries. It ships with the same 1M-token context window and 128k output ceiling as its bigger siblings, plus adaptive thinking controlled by the effort parameter.

Cowork everywhere, and Duet goes proactive

The model news is great, but the product changes are where your day-to-day actually shifts.

Cowork in every conversation. Before September, you picked a mode up front. Now Cowork's capabilities are available from any chat, and your chats, tasks, projects, connectors, and skills carry over. Ask a quick question, then say "okay, now build the spreadsheet," and it just keeps going. It's rolling out gradually to Pro and Max on web, desktop, and mobile.

Design, Slides, and Docs for everyone. You can ask for a design, a deck, or a document in any conversation, including in Claude Code, and then edit it with Claude or by hand. These are on every plan, including Free. (On Enterprise they're in beta and off until an owner turns them on.)

Claude Duet. This is the big swing. Duet is a new app where Claude acts as a proactive agent: rather than waiting for you to type, it works alongside you and moves things forward. It's a research preview for Pro and Max users, so expect rough edges. But if you've ever wished your assistant would notice the follow-up you forgot, this is Anthropic's answer.

Plugins. Developers can now submit plugins to the Claude directory through a new developer portal, track them through review, and see usage analytics once they're live. If you build tools for a niche (security, legal, data), this is a new distribution channel.

A Claude workflow you can steal today

New features only matter if they change how you work. Here's the loop we use for research-heavy articles and security work. It leans on the new defaults.

1. Scope in one message. Tell Claude the goal, the audience, and what "done" looks like. Be specific about constraints, for example "no exploit code" or "cite primary sources only."
2. Let it research, then verify the claims that matter. Ask for a short list of the facts it's relying on, with sources. Spot-check the ones your conclusion depends on. Models are good; they're not a substitute for checking the primary source.
3. Draft in a Doc, not a chat bubble. Ask for the output as a document so you can edit it in place and come back to it.
4. Hand repetitive work to Haiku 5.5. If you're tagging, extracting, or summarizing hundreds of items, a small fast model is cheaper and quicker. Save Opus 5.5 for judgment calls.
5. Make it repeatable. Once a workflow works, turn it into a skill or a project with standing instructions so you aren't re-explaining it every time.

A starter prompt for step 1:

Goal: a 1,200-word explainer on <topic> for <audience>.
Done means: accurate, plain English, every factual claim traceable
to a primary source, no speculation presented as fact.
Constraints: <anything off-limits>.
First, list the 5–10 facts you'll rely on and where each comes from.
Don't draft until I confirm the list.

That "list your facts first" step is the single biggest quality upgrade we've found. It turns a confident-sounding draft into one you can actually stand behind.

The best way to learn Claude right now

You can absolutely learn by poking at the app. But a structured course gets you to "I know why this works" much faster, and some of the best ones are free.

Start free: Anthropic Academy

Anthropic runs its own course platform, Anthropic Academy, at anthropic.skilljar.com. Every course is free, self-paced, needs only an email to sign up, and ends with a certificate. The catalog keeps growing, so check the site for the current list. The ones we'd start with:

• Claude 101. Everyday use of Claude and its core features. The right first stop if you're new or you've only used it for one-off questions.
• AI Fluency: Framework & Foundations. How to work with AI well: delegating, describing tasks, evaluating output. There are also tracks for educators, students, and nonprofits.
• Claude Code in Action. For developers: using Claude Code to read files, run commands, and modify code in a real repository.

These are the fastest zero-cost path from "I use Claude sometimes" to "I use Claude on purpose."

Go deeper: Building with the Claude API on Coursera

If you want to build with Claude (apps, agents, internal tools), Anthropic's Building with the Claude API on Coursera is the most complete option. It covers tool use, RAG, prompt evaluation, MCP, and agent workflows across seven modules. We wrote a full review of Building with the Claude API, and it's our top pick for developers.

Which path is right for you?

The bottom line

This fall Claude got smarter, cheaper, and more present. Opus 5.5 brings near-top-tier performance to the default experience. Haiku 5.5 makes high-volume AI work affordable. Cowork and Duet push Claude from "a chat window you visit" toward "a collaborator that keeps work moving."

The people who'll get the most out of it aren't the ones with the fanciest prompts. They're the ones with a repeatable workflow and enough grounding to know when to trust the output. Spend an afternoon on the free Anthropic Academy courses, steal the workflow above, and you'll be ahead of most of the people talking about AI this year.

Want more? Browse everything in AI & Workflows, or see how Claude stacks up in Claude vs ChatGPT vs Gemini.