Claude AI just got a lot more confusing (here's the 2026 lineup explained)

If you learned Claude as "pick Haiku for cheap, Sonnet for balance, Opus for hard stuff," that mental model is out of date. Anthropic didn't just update the three models — it added an entirely new tier sitting above Opus, renamed and relaunched half the lineup, and changed how the free chat interface routes your prompts behind the scenes. Most people using Claude right now are still working off outdated assumptions.

This guide breaks down exactly what changed, what each model is actually good for, and how to set up Claude so you're not overpaying for power you don't need (or underpowering a task that deserves better).

What actually changed in Claude's model lineup?

As of July 2026, Anthropic's generally available lineup looks like this: Claude Fable 5 as the top tier above Opus, Claude Opus 4.8, Claude Sonnet 5 released June 30, 2026 and now the default for Free and Pro, and Claude Haiku 4.5.

The biggest structural change is Fable 5, which introduced something new entirely — a "Mythos-class" tier that sits above Opus in raw capability. The Claude family now covers three named size tiers — Opus, Sonnet, and Haiku — joined in June 2026 by the Mythos-class tier that sits above the Opus class, with the current frontier being Claude Fable 5. There's also a sibling model called Mythos 5 that shares the same raw capability but with fewer safety guardrails — it's offered only in limited availability through what Anthropic calls Project Glasswing. Unless you work at an enterprise with a direct Anthropic relationship, you'll never touch Mythos 5. Fable 5 is the version everyone else gets.

Fable 5 had a weird few weeks after launch, too. Less than a week after its release, Anthropic disclosed that a US government export-control directive forced it to suspend access to both Fable 5 and Mythos 5, before redeploying Fable 5 globally on July 1 with updated cybersecurity safeguards, the same day it released Claude Sonnet 5. If you tried Fable 5 in mid-June and found it missing, that's why.

Sonnet 5 is arguably the bigger deal for everyday users, though, because it's now the default model for both Free and Pro plans, meaning most people are already using it without realizing anything changed.

How do you sign up and get started with Claude?

Head to claude.ai and create an account with Google, Apple, or an email address — no credit card required for the free tier. Anthropic's own Get Started guide walks through the basics, and the advice holds up well: the best approach is to speak to Claude like you would a coworker or friend, naturally and conversationally, whether you're asking simple questions or making complex requests like multi-step projects, analysis, creative writing, coding, or technical tasks.

Once you're in, you can see and switch which model you're talking to. The model you're currently chatting with is displayed either below your text input on web and desktop, or at the top of the screen on mobile. That matters more now than it used to, because the gap between models is wider than it used to be.

If you want a deeper walkthrough of features beyond basic chat — document uploads, search, extended thinking, and research mode — Anthropic's Getting Started with Claude.ai tutorial covers all of it directly from the source.

Which Claude model should you actually use?

This is the question everyone's getting wrong right now, because the old "Sonnet is the safe middle choice" advice needs an asterisk in 2026.

Sonnet 5 — your default for almost everything. This is the model to leave selected for 90% of your work. It's not just cheaper Opus anymore; independent benchmarking shows Sonnet 5 scoring 63.2% on SWE-bench Pro against Opus 4.8's 69.2%, while slightly outperforming Opus 4.8 on knowledge work tasks. Anthropic itself describes it this way: its performance is close to that of Opus 4.8, but at lower prices. For coding, writing, research, and daily tasks, this is the one to reach for first.

Opus 4.8 — save it for the genuinely hard stuff. Reserve this for accuracy-critical work: complex multi-file refactors, long autonomous agent runs, or reasoning tasks where you can't afford a wrong answer. A practical routing policy follows: send most agentic coding, tool use, and knowledge work to Sonnet 5, reserve Opus 4.8 for accuracy-critical tasks, and keep Haiku 4.5 for high-volume, latency-sensitive calls.

Haiku 4.5 — for scale, not depth. This is the model for volume. Haiku 4.5 is the fastest current Claude model and the cheapest, and it's more capable than people expect — Anthropic frames it as near-frontier intelligence — earning its place in high-volume classification, real-time chat surfaces, and background AI features inside production apps. Think customer support ticket triage or summarizing thousands of log entries — jobs where speed and cost matter more than nuance.

Fable 5 — only when the task justifies it. This is the model for the rare session where you genuinely need the most capability available, and the pricing reflects that. Fable 5 is the most powerful and most intelligent model Anthropic offers, a new tier that sits above Opus. Most people won't need it day-to-day, but for a week-long research project or a gnarly technical problem, it's worth knowing it exists.

One more practical note: don't assume the free version of Haiku is fully current on world events. Its reliable knowledge cutoff is Feb 2025 — older than Sonnet 5 and Opus, both at Jan 2026 — so for questions about recent events, Haiku may have stale knowledge.

Why does agentic research change everything?

If you remember Claude's old "deep research" feature as a glorified search summarizer, forget that. The current version plans its own investigation. It maps out a research strategy, runs multiple searches that build on each other's findings, cross-references sources, and returns a structured, cited report — closer to how a human analyst would tackle an open-ended question than a single search-and-summarize pass.

This matters because it changes what you should even ask Claude to do. Instead of "summarize this topic," you can now hand it a genuinely open-ended brief — "what are the tradeoffs between these three market entry strategies" — and let it figure out the research plan itself. Give it room to work; complex agentic research can take several minutes rather than seconds, and that's the tradeoff for depth.

How should you use Claude for coding now?

Sonnet 5's self-correction is the real story here. Instead of just generating code and stopping, it's now reasoning about the broader context of what it built — catching logical dead-ends, awkward edge cases, or design decisions that would eventually break something, often before you even flag the issue.

For actual development work, Claude Code is the tool to install rather than pasting code back and forth in the chat window. Claude Code is Anthropic's AI coding CLI that works directly with your files and terminal, integrating into your development environment instead of requiring copy-paste workflows. If you're a developer weighing model choice for real production code, MarkTechPost's benchmark comparison of Sonnet 5, Sonnet 4.6, and Opus 4.8 is worth reading before you commit to a workflow, since Anthropic frames this release around agentic reliability rather than one headline benchmark — longer task chains without losing context, better self-correction when a tool call fails, and steadier behavior across extended sessions inside Claude Code or Cowork.

For high-volume, repetitive coding tasks — batch file edits, routine reviews, quick classification of code changes — spinning up multiple Haiku sub-agents and merging their output back into one clean file is a genuinely underused pattern. It's fast and cheap, and Haiku's speed advantage makes it well suited to exactly this kind of parallel, low-stakes work.

What should beginners do this week?

Don't overthink the model selector. Start every conversation on Sonnet 5 — it's the default for a reason, and Anthropic's own pricing structure backs that up: introductory pricing for Sonnet 5 runs $2 per million input tokens and $10 per million output tokens through August 31, 2026, before moving to standard $3/$15 pricing. Only reach for Opus when a task genuinely stalls out or needs deeper reasoning, and only reach for Fable 5 when the stakes or complexity of the work justify a completely different price tier.

Read Anthropic's models overview documentation once, even if you never touch the API — it's the clearest explanation of what each tier is built for, straight from the source. Pair that with the official Claude Help Center for interface basics, and you'll understand this lineup better than most people who've been using Claude for years.

The model landscape will keep shifting — that's just the pace Anthropic is shipping at right now. But the underlying skill stays the same: matching the size of the tool to the size of the problem. Master that, and every future model release becomes easy to slot into a workflow you already understand.