Which Claude Model Should You Use in 2026? Opus vs Sonnet vs Haiku vs Fable

Anthropic's Claude lineup is no longer a single model with a slider. In 2026 there are four models you can use today, each tuned for a different point on the speed-cost-capability curve: Haiku, Sonnet, Opus, and Fable. Fable 5 briefly disappeared in June after a government-directed suspension, but it was restored on July 1, 2026 (more on that below). If you are calling Claude through the API, building an app, or just trying to understand what powers a given plan, the question "which Claude model should I use" has a real answer, and it depends almost entirely on the task in front of you.

This guide compares the Claude models head to head and gives you a simple way to choose. If you are brand new to Claude, our free Claude for Beginners course is the fastest way to get comfortable before you start picking models.

A quick note on scope: this post is about the models, not the consumer subscription plans. If your real question is "which Claude plan should I buy" (free vs Pro vs Max), head to our free vs Pro vs Max comparison instead. Here, we focus on choosing the right model for a task.

The Claude Model Lineup at a Glance

Here is the full lineup, from most cost-effective to most capable. The prices below are API list prices per million tokens as of publication, so treat them as a snapshot rather than a permanent quote. The relative positioning is what matters most.

Model	API model ID	Context window	Max output	Input / output (per 1M tokens)	Status / best for
Claude Haiku 4.5	claude-haiku-4-5	200K	64K	$1 / $5	Available. Fast, cheap, high-volume simple tasks
Claude Sonnet 5	claude-sonnet-5	1M	128K	$3 / $15 ($2 / $10 introductory through August 31, 2026)	Available. The new default and everyday workhorse
Claude Opus 4.8	claude-opus-4-8	1M	128K	$5 / $25	Available. Hardest reasoning, deep coding, long-horizon agents
Claude Fable 5	claude-fable-5	1M	128K	$10 / $50	Available again (restored July 1, 2026). The frontier option

A few things stand out. Sonnet, Opus, and Fable all share a 1M-token context window, which is large enough to hold an entire codebase or a stack of long documents at once. Haiku trades that down to a still-generous 200K-token window in exchange for being the fastest and cheapest option. And the ladder now climbs cleanly: each step up buys more capability at a higher per-token price, with Fable 5 sitting at the top as the frontier model.

One migration note: Sonnet 5 replaced Sonnet 4.6 as the Sonnet tier and the default model on June 30, 2026. If you built a workload on Sonnet 4.6, plan to test it on Sonnet 5 and check Anthropic's model documentation for the older model's support timeline.

Claude Haiku 4.5: Fast and Cheap

Haiku is the model you reach for when you need speed and volume more than raw intelligence. It is the most cost-effective option in the lineup, and it is genuinely fast, which makes it ideal for tasks you run thousands of times: classifying support tickets, extracting fields from documents, tagging content, routing requests, or powering an autocomplete-style feature.

You should not use Haiku for tasks that need deep reasoning, careful multi-step planning, or the kind of nuanced writing where small mistakes compound. That is what the bigger models are for. But for the long tail of simple, repetitive work, Haiku does the job at a fraction of the cost, and its 200K-token context window is more than enough for most of these jobs.

Rule of thumb: if the task is simple and you are going to run it a lot, start with Haiku.

Claude Sonnet 5: The New Default

Sonnet 5, released June 30, 2026, is the model most people should default to, and it is now literally the default: it powers the standard experience for free and Pro users on claude.ai. It offers the best balance of speed, intelligence, and price in the lineup, and Anthropic positions it as closing much of the gap to Opus 4.8 on many tasks, including agentic work that used to require the bigger, pricier models.

That makes it the right choice for the bulk of real work: writing and editing, summarizing long documents, answering questions, drafting code, building chatbots, and running agents that plan, use tools, and work through multi-step tasks. Sonnet 5 carries the full 1M-token context window and a 128K max output, so you can feed it large inputs and get substantial results back without splitting the work up.

On price, note the introductory offer: through August 31, 2026 the API rate is $2 / $10 per million tokens (input / output), after which it moves to the standard $3 / $15. If you are modeling costs past that date, use the standard rate.

Rule of thumb: when you are not sure which model to pick, pick Sonnet 5. It is the workhorse, and it is right far more often than it is wrong.

Claude Opus 4.8: The Flagship for Hard Problems

Opus is where you go when the task is genuinely hard and accuracy matters more than cost. Anthropic positions it for state-of-the-art long-horizon agentic work, deep coding, knowledge work, and memory: the kind of tasks that run for many steps, span many files, or require the model to hold a complex goal in mind and pursue it autonomously. Even with Sonnet 5 closing the gap on many tasks, Opus remains the model of choice when you need the higher accuracy ceiling without stepping all the way up to Fable pricing.

Concrete cases where Opus earns its higher price over Sonnet:

• Deep, multi-file code refactors where the model has to understand a whole system before it changes anything.
• Long-horizon agentic loops that string together many tool calls and decisions toward one goal.
• Hard reasoning and analysis where getting the answer right matters more than getting it cheap.
• Knowledge work that benefits from the model planning carefully and verifying its own output.

With a 1M-token context window and a 128K max output, Opus has the room to take on big inputs and produce substantial results. Want to get more out of it on tasks like these? Our free Prompt Engineering course teaches the techniques that separate good prompts from great ones.

Rule of thumb: reach for Opus when Sonnet starts to struggle with the complexity, depth, or length of the task.

Claude Fable 5: Suspended, Then Restored

Fable 5 is Anthropic's most capable model, built for the most demanding reasoning and long-horizon agentic work. It sits above the Opus tier at $10 / $50 per million tokens (input / output), with a 1M-token context window and a 128K max output.

Its short life has already been eventful. It launched on June 9, 2026 as the new frontier option. On June 12, 2026, Anthropic suspended it (along with its sibling model, Mythos 5) following a US government directive issued under export-control authority. Then, on July 1, 2026, after the Department of Commerce lifted the order, Anthropic redeployed both models globally across the Claude API, the Claude apps, Claude Code, and Claude Cowork, together with an improved safety classifier addressing the jailbreak that triggered the directive.

The practical takeaway: you can use Fable 5 again, and it is once more the top of the ladder for tasks where you want the most capable model Anthropic offers, priced accordingly. How it is metered on consumer plans has been shifting since the restoration, so check your plan's current terms before leaning on it heavily there; through the API it is straightforward pay-per-token.

For the full story of the suspension and restoration, see our Claude Fable 5 suspended explainer.

How to Choose: A Simple Decision Guide

Strip away the names and the decision comes down to matching the task to the right tier, across four live models:

• Simple, cheap, high-volume task? Use Haiku 4.5.
• Balanced everyday work (writing, summarizing, most coding, chatbots, agents)? Use Sonnet 5 as your default.
• Hardest reasoning, deep multi-file coding, long-horizon agents where accuracy is critical? Step up to Opus 4.8.
• Frontier-level work where you want the most capable model available, and the budget to match? Reach for Fable 5.

A practical pattern many teams use: start with Sonnet for almost everything, drop down to Haiku for the simple high-volume parts of your workload, and step up to Opus or Fable only for the specific tasks that need them. That way you are paying for capability exactly where it earns its keep, and not a token more.

If you are also weighing Claude against other tools, our Claude vs ChatGPT vs Gemini comparison covers the broader picture, and if you want the equivalent breakdown for OpenAI, see which ChatGPT model to use.

A Note on Pricing and Plans

Two reminders before you commit. First, the dollar figures in the table are API list prices as of publication, and model pricing can change. Sonnet 5 in particular carries an introductory rate only through August 31, 2026, so always confirm the current numbers on Anthropic's pricing page before you build a cost model around them. The relative ordering (Haiku cheapest, Fable most expensive) is the durable takeaway.

Second, if you access Claude through a consumer subscription rather than the API, the "which model" question is wrapped up with "which plan." The plans determine which models you can reach and how much you can use them. For that side of the decision, our free vs Pro vs Max comparison is the place to go. This guide stays focused on the models themselves.

Key Takeaways

• Four Claude models are available to use today: Haiku 4.5 (fast and cheap), Sonnet 5 (the new default), Opus 4.8 (the flagship for hard problems), and Fable 5 (the frontier option, restored July 1, 2026).
• Sonnet 5 replaced Sonnet 4.6 as the default on June 30, 2026, with a 1M-token context window and performance close to Opus 4.8 on many tasks. Its introductory API pricing runs through August 31, 2026.
• Fable 5 is back. It was suspended on June 12, 2026 under a US government directive and restored on July 1, 2026 after the order was lifted.
• Match the model to the task: simple and high-volume goes to Haiku, everyday work to Sonnet, hard or long-horizon work to Opus, and frontier work to Fable.
• Sonnet 5, Opus 4.8, and Fable 5 all have a 1M-token context window; Haiku has 200K.
• Prices shown are API list prices as of publication and can change, so verify before building a cost model.

Ready to put this into practice? Start with our free Claude for Beginners course to get fluent with Claude, then sharpen your results with Prompt Engineering. Both are free and come with a shareable certificate.