Claude Models: Cost, Differences, & When to Use Each

Claude models are a family of generative AI models built by Anthropic. The current lineup runs from Claude Fable 5, the most intelligent model in the family, down through Claude Opus 4.8, Claude Sonnet 5, and Claude Haiku 4.5. Each tier trades some intelligence for more speed and lower cost.

This guide explains what each model does, how much it costs, and which one fits the task in front of you, whether you write essays, review data, or build software.

What Are the Claude Models?

Claude models are large language models that handle natural language conversation, writing, coding, and analysis. Like other large language models, they are trained on vast amounts of text and learn patterns in language from that data. They are built on the transformer architecture, a design that lets the model track relationships between words across very long passages. That design is why Claude can parse a full legal contract or research paper and answer detailed questions about any part of it.

Anthropic groups the models into tiers. Each intelligence category serves a different balance of capability, speed, and price. The smartest models cost more and respond more slowly. The fastest models cost less and handle simpler work. You pick based on what the task demands.

All current Claude models accept text input and images, and all of them work across many languages. They can read charts, technical diagrams, photos, and other visual formats, then answer questions about what they see. That matters if your work lives in PDFs, slides, or screenshots instead of plain text.

Read: The 5 Best AI Personal Assistants: Reviewed & Ranked (2026)

The Current Claude Model Lineup (2026)

Anthropic released Claude Fable 5 on June 9, 2026, adding a new top tier above Opus. Here is how the four main models compare on the Claude API today.

Sonnet 5 has introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026.

A context window is how much text a model can hold in memory at once. One million tokens cover roughly 750,000 words, enough for an entire codebase or a stack of research papers. Claude Haiku 4.5 is the cost-effective model of the group. It handles quick questions and high-volume work where rapid responses matter more than deep reasoning.

The top three models share a reliable knowledge cutoff of January 2026, so their built-in knowledge stays current through that date. Haiku's cutoff is February 2025.

One detail that trips up new users is versioning. Claude models a ship as pinned snapshots, which means a name like Opus 4.8 refers to one fixed release rather than a moving target. When Anthropic improves a model, it publishes a new version with a new number instead of quietly changing the old one. That gives developers stable behavior and gives you a simple way to tell whether an article or benchmark reflects the current generation.

How Claude Models Have Changed From Previous Models

The Opus, Sonnet, and Haiku names date back to March 2024, when Anthropic introduced the Claude 3 family. At release, Claude 3 Opus cost $15 per million input tokens and $75 per million output tokens with a 200K context window. It was also an early standout at long-context recall, scoring above 99 percent accuracy on needle-in-a-haystack tests that hide one fact inside a huge block of text. Compare that model with today's Opus 4.8 at a third of the price, five times the context, and far stronger performance. The pattern across every generation has been more capability at a lower cost.

The Claude 4 generation brought models like Sonnet 4.5, Opus 4.6, and Sonnet 4.6, each a step up in coding and agent skills. The June 2026 release of Claude Fable 5 changed the structure itself. For the first time, a tier sits above Opus. This matters when you read older comparisons. An article weighing Sonnet 4.5 against Opus 4.5 describes models that have since been replaced twice. Check the publish date on any Claude comparison before you act on it.

What Is Claude Mythos and Who Can Use It?

Claude Mythos 5 shares the same underlying model as Claude Fable 5 but ships without certain dual-use safety measures. It is not open to the public. Access is granted at Anthropic's discretion and prioritized for approved organizations doing defensive cybersecurity work. The earlier Claude Mythos Preview follows the same rules. Neither model is in public preview, and there is no self-serve sign-up. Organizations that believe they qualify can contact their Anthropic, AWS, or Google Cloud account team to request access. If you see the Mythos name in coverage, know that Fable 5 is the version you can actually use.

Key Differences Between Claude Opus and Claude Sonnet

Claude Opus is the stronger choice for complex tasks that need deep, multi-step thinking. Claude Sonnet runs faster, costs less, and performs at near-human levels on most everyday work. If you are unsure where to start, start with Sonnet and move up only when a task proves too hard for it.

Intelligence and Reasoning

Opus 4.8 leads the family in complex reasoning, code generation, and work that spans many steps. It can reason over an entire codebase, hold context across a multi-day project, and process high-resolution images. Anthropic positions it for coding, financial analysis, cybersecurity, and computer use, which lets the model operate software the way a person would.

Sonnet trails Opus on the hardest problems but stays competitive with other leading models on general intelligence benchmarks while responding faster. For a concrete example, take two writing tasks. Editing a two-page cover letter is comfortable Sonnet territory. You get a polished draft in seconds. Restructuring a 40-page thesis with tangled arguments is where Opus earns its price, because it tracks how a change on page 3 affects a claim on page 31. The gap shows on tasks with many moving parts, not on tasks with one clear job.

Independent testing backs this up. When the team at Cosmic built the same application with Sonnet 4.5 and Opus 4.5 from a single prompt, both produced working software, but Opus made sharper architecture decisions and anticipated needs the prompt never stated. Better results came from the model's judgment, not from extra instructions.

Read: Claude vs. ChatGPT vs. Gemini: Pros & Cons and Which AI Tool is Best for You

Speed and Cost

Sonnet costs less per token and returns faster response times, which makes it the default for high-volume work. Opus complicates the math through token efficiency. A smarter model often finishes the same job with fewer words. In that same Cosmic test, Opus used about 19 percent fewer total tokens than Sonnet to build a comparable app. Fewer input tokens and less output per task shrink the real price gap between the two tiers. If you run thousands of requests, measure cost per completed task rather than cost per token.

Effort Control and Thinking Modes

Newer Claude models give developers effort control, a setting that trades answer quality against cost and speed. You can set the effort level to low, medium, or high, and Opus models also support a max setting for the hardest problems. Anthropic reported that Opus at medium effort matched the best Sonnet coding benchmark score of its generation while using 76 percent fewer output tokens.

The models also support adaptive thinking. Instead of reasoning step by step on every request, the model decides when a question deserves extended thought and when a direct answer works. Together, these tools blur the old line between tiers. An Opus model dialed down can act like a fast model. A Sonnet model on a hard prompt can slow down and think. The tier you pick sets the ceiling. The settings decide how much of that ceiling you use.

What Is Claude Fable 5?

Claude Fable 5 is Anthropic's most intelligent model and the first in the new Mythos-class tier above Opus. Released in June 2026, it targets agentic AI work, meaning tasks where the model plans, acts, and checks its own output over long stretches instead of answering one prompt at a time.

Fable 5 holds a 1M-token context window and applies deep contextual understanding across everything in it. It can reason over a full codebase, coordinate sub-agents that each handle part of a project, and verify its own work as it goes. Anthropic describes it as capable of longer independent work than any prior Claude model. It also carries extra safety classifiers that can refuse requests touching dual-use areas like advanced biology or offensive security, a design choice tied to how capable the model is.

At $10 per million input tokens and $50 per million output tokens, Fable 5 costs double Opus 4.8. That premium makes sense for large migrations, multi-day autonomous coding, and vision-heavy agent workflows. It does not make sense for drafting emails or summarizing articles. If Opus already handles your task well, the state-of-the-art tier adds cost without adding value that you can see.

Read: Agentic AI vs. AI Agents: Differences & What You Need to Know

Which Claude Model Should You Use?

Match the model to the task, not the other way around. There is no single best model, only the best model for a given job. Most people get strong results from Claude models by using Sonnet as their default, dropping to Haiku for quick or repetitive jobs, and stepping up to Opus or Fable only when a task keeps failing at the lower tier.

Task-to-Model Decision Table

One more point on defaults. The gap between tiers shows up at the edges of difficulty. For the middle 80 percent of tasks, model choice matters less than a clear prompt. Say what you want, give an example of good output, and state the format you need. A well-prompted Sonnet beats a vaguely prompted Opus most days.

Read: 20 Examples of AI Agents and Workflows: Real Use Cases by Business Function

For Students and Non-Developers

You do not need the Claude API to use these models. Visit claude.ai to chat with Claude directly in your browser or the mobile app. The app gives you a model picker in a dropdown, and the free plan includes access to a current model with usage limits. Paid plans raise those limits and open the top tiers.

For application prep and coursework, Sonnet covers almost everything. It can critique a personal statement, run mock interview questions, explain a concept you missed in lecture, and turn dense readings into study notes. The one habit that matters more than model choice is verification. Every generative AI model can produce incorrect information with full confidence, including fake citations and wrong dates. Treat Claude as a fast first draft of knowledge, then confirm anything you plan to submit, cite, or say in an interview against a primary source. Admissions committees and hiring managers judge your accuracy, and the model will not sit in the room with you.

Read: Claude vs. ChatGPT vs. Gemini: Pros & Cons and Which AI Tool is Best for You

For Developers Using Claude Code, the Claude API, and Tool Use

Developers reach Claude models through several routes. The Claude API gives you direct programmatic access with a search tool, file handling, and structured output. You can also query model capabilities and token limits programmatically through the Models API, which returns limits and features for every available model. Output limits depend on how you call the model. The standard cap on current top models is 128K tokens per response, and Opus 4.8 can produce up to 300K output tokens on the Message Batches API with a beta header enabled.

Claude Code brings the models into your terminal for agentic coding, where the model reads your repository, writes changes, runs tests, and iterates on failures. It fits well when you want to create features or fix bugs without hand-holding every step.

Tool use lets the models call outside functions, query data sources, and take actions in other software. Computer use extends to operating a screen directly. These capabilities power the current wave of agents in the enterprise world, from research assistants to workflow automation.

If your company runs on a specific cloud, availability is broad. Claude models ship through the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Foundry currently offers the lineup in preview, from Haiku up through Fable 5, with quotas that vary by subscription. Whichever platform you choose, you get the same underlying model generation.

Read: Claude Code Subagents vs. Agents: Differences & How to Use Each Effectively

How Much Do Claude Models Cost?

API pricing spans $1 to $10 per million input tokens and $5 to $50 per million output tokens across the current lineup. Here is the full table one more time in plain numbers.

To make those numbers concrete, a million tokens equals roughly 750,000 words. Asking Sonnet to review a 5,000-word essay and return detailed feedback might use around 10,000 tokens total, which costs a few cents. Costs only become a serious line item at high volume or with the top tiers on long agent runs.

Most individual users never touch API pricing. On claude.ai, you pay a flat subscription instead, and the same Claude models power both routes. The free tier works for light use. Pro and Max plans raise limits and add the stronger models, and Team plans cover groups. If you send fewer than a few dozen substantial requests per day, a subscription is simpler and usually cheaper than metered API billing.

You should check current plan pricing on Anthropic's site before you commit, because plans change more often than model prices do.

How Safe Are Claude Models?

Anthropic builds safety into the models rather than bolting it on afterward. The core method is Constitutional AI, a training approach that teaches the model to follow a written set of principles so it stays helpful and honest while avoiding harmful output. Anthropic also reported hardened resistance to prompt injection, a type of attack that hides malicious instructions inside content the model reads, starting with the Opus 4.5 generation.

The company governs releases under its Responsible Scaling Policy, which assigns AI Safety Levels based on model capability. Anthropic's most capable current models, including Claude Fable 5, ship under ASL-3 safeguards. Those include real-time classifiers that screen requests in sensitive areas like biology and cybersecurity, stronger security around the models themselves, and rapid response to jailbreak attempts. For you as a user, the practical effect is occasional refusals of requests near those areas. For most schoolwork, writing, and career tasks, you will never notice the safeguards at all.

The Bottom Line

The right Claude model saves you hours on drafts, research, and prep. What it cannot do is tell you how an admissions reader will react to your story or how an interviewer will hear your answers. That judgment comes from people who have sat on the other side of the table.

Get More From the AI Tools You Use

Browse Leland's coaches to pair fast AI output with expert human feedback on your applications and career moves. Our AI expert coaches can help you do that. They review your essays, run mock interviews, and pressure-test the work you produce with AI, so it holds up in front of a real decision-maker. If you want to go deeper on the AI side, enroll in Leland's AI Builder Program, where you learn to build with tools like Claude in a structured, hands-on course. You can also browse our live programs to find upcoming sessions on admissions, careers, and AI skills.

Start with a coach, a program, or both, and turn what you learned in this guide into results!

Keep learning about AI tools:

• The 5 Best AI Coding Agents: Pros & Cons, Reviews, & Which is Best for You
• Context Engineering: What it Is & Why It's Important for Your AI Usage (2026)
• How to Build an AI App: Best Process, Tools, & Tips (2026)
• MCP: What It Is, Protocol, & Everything You Need to Know
• The 5 Best AI Tools & Agents for Productivity: Reviewed & Ranked (2026)

FAQ

What are the different Claude models?

• The current Claude models are Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5, and Claude Haiku 4.5. Fable 5 is the most intelligent model and handles long-running agent work. Opus 4.8 covers complex reasoning and coding. Sonnet balances speed and quality for everyday tasks, and Haiku is the fastest, cheapest option. Anthropic also offers Claude Mythos 5, a gated release limited to approved organizations.

How do you switch models in Claude Code?

• Run the /model command inside a Claude Code session. Typing /model alone opens an interactive picker, or you can name the model directly, such as /model sonnet or /model claude-opus-4-8. The switch takes effect immediately, and your conversation context carries over. Run /status anytime to check which model is active. Selecting Fable 5 requires Claude Code version 2.1.170 or later.

How do you change the default model in Claude Code?

• You have three options beyond in-session switching. Launch with the --model flag, as in claude --model opus, to set a model for one session. Set the ANTHROPIC_MODEL environment variable to make a model the default for every session from that shell. Or set the model field in your Claude Code settings file for a permanent default. In recent versions, picking a model with /model also saves it as your default for new sessions.

Is Claude Opus better than Sonnet?

• Opus is better at complex reasoning, hard-coding problems, and analysis with many steps. Sonnet is faster, cheaper, and strong enough for most writing, research, and everyday tasks. Pick Opus when a task keeps failing on Sonnet, not by default.

Which Claude model is free?

• The free plan on claude.ai includes access to a current Claude model with daily usage limits. Paid plans raise those limits and add the higher tiers. You do not need a developer account or API key to use the free version.