The four Claude models and where each fits
The four tiers and their roles (as of July 2026)
Claude is the conversational AI service built by Anthropic. The names come from literary forms: Haiku is short and light, Sonnet sits in the balanced middle, Opus is a major work, and Fable tops the range.
Capability, speed, and price differ sharply between tiers, so matching the model to the weight of the task is the core skill in using Claude well. The same request costs ten times as much on Fable 5 as on Haiku 4.5.
What each tier is good at
Here is each model in one line, following Anthropic's own descriptions.
- Fable 5: the newest and most capable model, built for long, complex tasks. It plans the steps itself and checks its own work, which also means it thinks longer and responds more slowly than the rest.
- Opus 4.8: for complex agentic coding and enterprise work — problems that genuinely need sustained thinking.
- Sonnet 5: the best combination of speed and intelligence, covering most day-to-day coding, writing, analysis, and research.
- Haiku 4.5: the fastest model with near-frontier intelligence, suited to instant answers, summaries, and simple extraction.
| Description | Next-generation intelligence for long-running agents | For complex agentic coding and enterprise work | The best combination of speed and intelligence | The fastest model with near-frontier intelligence | — Models overview, "Latest models comparison" table (Fable 5 / Opus 4.8 / Sonnet 5 / Haiku 4.5, in order)
There is also the invitation-only Mythos 5 (through Project Glasswing) and legacy models such as Opus 4.7, 4.6, Sonnet 4.6, and 4.5, which remain callable through the API. For most people, though, the choice is between the four above. For the basics of the service itself, see "What Is Claude? Pricing, Choosing a Model, and Claude vs ChatGPT".
Spec comparison as of July 2026
Here are the headline specs from the official documentation.
| Item | Fable 5 | Opus 4.8 | Sonnet 5 | Haiku 4.5 |
|---|---|---|---|---|
| API input price | $10/MTok | $5/MTok | $2/MTok (introductory) | $1/MTok |
| API output price | $50/MTok | $25/MTok | $10/MTok (introductory) | $5/MTok |
| Context window | 1M tokens | 1M tokens | 1M tokens | 200k tokens |
| Max output | 128k tokens | 128k tokens | 128k tokens | 64k tokens |
| Comparative latency | Slower | Moderate | Fast | Fastest |
| Reliable knowledge cutoff | Jan 2026 | Jan 2026 | Jan 2026 | Feb 2025 |
| Adaptive thinking | Yes (always on) | Yes | Yes | No |
| Extended thinking | No | No | No | Yes |
| Chat plans that include it | Pro and above | Pro and above | Free and above | Free and above |
MTok = million tokens. The Sonnet 5 introductory rate runs through August 31, 2026; from September 1 it becomes $3 input / $15 output.
| Feature | Claude Fable 5 | Claude Opus 4.8 | Claude Sonnet 5 | Claude Haiku 4.5 | | Pricing | $10 / input MTok $50 / output MTok | $5 / input MTok $25 / output MTok | $3 / input MTok $15 / output MTok | $1 / input MTok $5 / output MTok | | Context window | 1M tokens | 1M tokens | 1M tokens | 200k tokens | | Max output | 128k tokens | 128k tokens | 128k tokens | 64k tokens | | Reliable knowledge cutoff | Jan 2026 | Jan 2026 | Jan 2026 | Feb 2025 | — Models overview, "Latest models comparison" table
Fable 5, Opus 4.8, and Sonnet 5 all take a 1M-token context window, so Haiku 4.5 is the only one limited to 200k when you want to load a long document in one go. Max output lines up the same way: 128k for the top three, 64k for Haiku.
"Reliable knowledge cutoff" is the date through which the model's knowledge is most extensive and reliable. The top three sit at January 2026, but Haiku 4.5 stops at February 2025 — nearly a year earlier. Research about recent events can outrun what Haiku knows.
One more thing the price table does not show. Fable 5, Opus 4.7 and later, and Sonnet 5 use a newer tokenizer that produces roughly 30% more tokens for the same text. Even at an identical per-token rate, the invoice for the same document comes out higher than on older models.
Claude Opus 4.7 and later Opus models, Claude Fable 5, Claude Mythos 5, Claude Mythos Preview, and Claude Sonnet 5 use a newer tokenizer that contributes to their improved performance on a wide range of tasks. This tokenizer produces approximately 30% more tokens for the same text. — Pricing, note under "Model pricing"
How Fable 5 differs from Opus 4.8
Fable 5 is not simply a faster Opus. The documentation describes Opus 4.8 as "For complex agentic coding and enterprise work" and Fable 5 as "Next-generation intelligence for long-running agents" — a step up in how long the work is expected to run.
The clearest practical difference is how much supervision it needs. Fable 5 is designed to work through a task with fewer mid-task check-ins: describe the outcome, and it plans the steps and reviews its own output. In exchange it thinks longer, responds the slowest of the four, and consumes the most of your rate limit.
Fable 5 is our newest and most capable model, built for long, complex tasks. It can work through tasks more autonomously with fewer mid-task check-ins — describe the outcome you want, and it plans the steps and checks its own work along the way. It takes time to think through problems before answering, so responses take longer, and it uses the most of your rate limit. — Choosing the right Claude model, "Meet the four models"
API pricing is also double that of Opus 4.8, at $10 input and $50 output per million tokens. Treating Fable 5 as the tier you move to when Opus 4.8 falls short keeps monthly spend predictable.
On the technical side, the difference shows up in how thinking is handled. Fable 5, Opus 4.8, and Sonnet 5 all support adaptive thinking, which adjusts reasoning depth to the task, and on Fable 5 it is always on and cannot be turned off. Haiku 4.5 goes the other way: no adaptive thinking, but it does support extended thinking, where you explicitly switch on step-by-step reasoning before the answer.
API pricing and plan comparison
API input price per million tokens (USD)
There are two ways to pay for Claude: the API, billed per token, and a subscription plan that gives you the chat interface at claude.ai for a fixed monthly fee. Software integrations go through the API; everyday chat use goes through a plan.
API price list (per million tokens)
Pricing splits into input and output tokens, with output always the more expensive side. Across every model the ratio is a consistent 1:5.
| Model | Input | Output |
|---|---|---|
| Fable 5 | $10 | $50 |
| Opus 4.8 | $5 | $25 |
| Sonnet 5 (through August 31, 2026) | $2 | $10 |
| Sonnet 5 (from September 1, 2026) | $3 | $15 |
| Haiku 4.5 | $1 | $5 |
Introductory pricing of $2/$10 per million input/output tokens is in effect through August 31, 2026, after which the standard pricing of $3/$15 per million input/output tokens will take effect. — Pricing, "Claude Sonnet 5 introductory pricing" note
Haiku 4.5 costs a tenth of Fable 5, and while the introductory rate lasts, Sonnet 5 costs a fifth. The more volume you push through, the more directly that gap lands on the invoice.
To put numbers on it, take a workload of 100k input and 50k output tokens per day. At the Sonnet 5 introductory rate that is $0.70 a day ($0.20 input plus $0.50 output), or about $21 over 30 days. The same volume on Opus 4.8 comes to roughly $52.50 a month, and on Fable 5 about $105. The API bills only what you use, so light usage can undercut a subscription — but heavy usage scales in proportion.
Legacy Opus models 4.7, 4.6, and 4.5 remain available through the API at the same $5 / $25 as Opus 4.8. Opus 4.1 is the exception at $15 / $75, and it is scheduled to be retired on August 5, 2026.
Claude Opus 4.1 (claude-opus-4-1-20250805) is deprecated and will be retired on August 5, 2026. Migrate to Claude Opus 4.8 before the retirement date. — Models overview, warning in the "Legacy models" section
What Free, Pro, and Max each include
In the chat interface at claude.ai, the plan you are on determines which models appear in the picker. This is the part that changed during 2026: the free plan no longer reaches the top of the range.
| Plan | Price | Models included |
|---|---|---|
| Free | $0 | Haiku, Sonnet |
| Pro | $17/month billed annually ($200 up front), $20 billed monthly | Haiku, Sonnet, plus Opus and Fable |
| Max | From $100/month | Same models as Pro, with substantially higher usage |
| Team | $20/seat/month billed annually, $25 billed monthly | Standard seats; premium seats are $100/seat annually, $125 monthly |
Which models you have and how high your limit is both depend on your Claude account plan. Free includes Haiku and Sonnet; Pro and Max adds Opus, Fable, and more headroom. — Choosing the right Claude model, introduction
Sonnet 5 — the everyday workhorse — is available on the free plan, so a lot of real work fits inside $0. Trying Opus 4.8 or Fable 5, however, requires Pro or above. Pro drops to an effective $17 a month when billed annually, so anything beyond a few months of use favours the annual option.
Max starts at $100 a month and offers the same model list as Pro with a much higher usage ceiling. For groups there is Team, and organisations that need SSO, audit logs, and custom data retention move to Enterprise.
Cutting costs with the Batch API and prompt caching
If API volume is high, two features bring the bill down substantially.
The Batch API takes requests that do not need an immediate response and processes them asynchronously, at a 50% discount on both input and output.
| Model | Batch input | Batch output |
|---|---|---|
| Fable 5 | $5 | $25 |
| Opus 4.8 | $2.50 | $12.50 |
| Sonnet 5 (through August 31, 2026) | $1 | $5 |
| Haiku 4.5 | $0.50 | $2.50 |
The Batch API allows asynchronous processing of large volumes of requests with a 50% discount on both input and output tokens. — Pricing, "Batch processing"
Prompt caching keeps a reusable system prompt or reference document in cache so that repeat requests are charged at 10% of the standard input rate.
| Operation | Multiplier | Duration |
|---|---|---|
| 5-minute cache write | 1.25× base input price | 5 minutes |
| 1-hour cache write | 2× base input price | 1 hour |
| Cache read (hit) | 0.1× base input price | Same as the preceding write |
A cache hit costs 10% of the standard input price, which means caching pays off after just one cache read for the 5-minute duration (1.25x write), or after two cache reads for the 1-hour duration (2x write). — Pricing, "Prompt caching"
With the five-minute cache, the write costs 1.25× the input rate, so a single cache hit already pays for it. Reusing a 10,000-token system prompt across 100 requests means full price every time without caching, and a 90% cut on input from the second request onward with it.
The two stack, so routing bulk document work through Batch plus caching brings the effective rate down sharply even on the upper tiers. Setup details are covered on the official Pricing page.
Choosing a Claude model by task
Step up as the work gets heavier
Specs and prices alone rarely settle the question. This section maps common jobs to the model that fits.
Recommended model by use case
| Use case | Recommended model | Why |
|---|---|---|
| Everyday chat and lookups | Sonnet 5 | Fast, inexpensive, and accurate enough |
| Coding and debugging | Sonnet 5 → Opus 4.8 | Start on Sonnet; escalate for tangled bugs and design calls |
| Reading and summarising long documents | Sonnet 5 | 1M-token context with fast throughput |
| High-volume routine work (classification, extraction) | Haiku 4.5 | Cheapest and fastest; the gap compounds with volume |
| Work that depends on recent events | Sonnet 5 / Opus 4.8 | Knowledge is reliable through January 2026 |
| Long autonomous jobs | Fable 5 | Plans and self-checks with fewer interruptions |
Sonnet 5 is the first candidate for most use cases. Anthropic's own tutorial tells readers to start there when unsure, and beyond coding, writing, analysis, and research it also handles computer use, vision tasks, and document and spreadsheet creation.
Sonnet is the daily driver. Sonnet 5 brings strong reasoning to the kind of work you do every day — coding, writing, analysis, research, and complex problem-solving. It's responsive enough for real-time collaboration and capable enough that most problems won't outgrow it. — Choosing the right Claude model, "Meet the four models"
In the chat interface you can also tune the effort setting, which controls how deeply the model thinks. On Opus 4.8 it defaults to high everywhere, including claude.ai, so dialling it down for light questions and up for involved analysis shifts the speed-versus-accuracy balance without changing model.
On Claude Opus 4.8, the effort parameter defaults to high on all surfaces, including the Claude API, Claude Code, and claude.ai. On Claude Sonnet 5, it defaults to high on the Claude API and Claude Code. — Models overview, note
Legacy models and retirement dates
If you have been using Claude for a while, older model names still show up in articles and documentation. Here is where they stand as of July 2026.
- Sonnet 4.6: moved to the legacy list when Sonnet 5 arrived. Still callable through the API at $3 input / $15 output.
- Opus 4.7, 4.6, 4.5: also legacy, but priced identically to Opus 4.8. There is little reason to pick them for new work, so consolidate on 4.8 where you can.
- Opus 4.1: deprecated, and retiring on August 5, 2026. Migrate any system still calling it to Opus 4.8 well before that date.
Every Claude API model ID is a pinned snapshot. Even IDs without a date in them do not silently roll forward to a newer release, so upgrading always means changing the ID explicitly.
Opus 4.7 and 4.8 also offer a "fast mode" for latency-sensitive work. On Opus 4.8 it is priced at $10 input / $50 output — double the standard rate. For a closer look at the cost differences between Opus versions, see "Claude Opus 4.8, 4.7, 4.6 Pricing Compared: Cost Differences and How to Choose".
Mixing models rather than picking one
In practice the efficient approach is not to settle on a single model but to switch as the work changes weight. A three-step pattern works well.
- Light work (email drafts, short questions, routine processing) → Haiku 4.5: instant responses and the smallest draw on your rate limit.
- Normal work (writing, code review, research) → Sonnet 5: the best balance of speed and accuracy.
- Heavy work (complex bug fixes, design decisions, long autonomous runs) → Opus 4.8 / Fable 5: accuracy improves where deep reasoning is genuinely needed.
Switching is a model-picker change on claude.ai and a model ID change in an API request. You can switch mid-conversation, which makes "try the light model first, escalate if it falls short" a practical default.
For automated pipelines, you can have Haiku 4.5 triage incoming requests and route only the hard ones to Sonnet 5 or Opus 4.8 — spending on the upper tiers exactly where accuracy is required.
Developers also have Claude Code, a CLI tool for driving Claude from the command line. Because it combines effort settings with model switching, light questions can run on Sonnet at low effort while design decisions run on Opus at high effort, all inside one environment.
If you are comparing across providers, ChatGPT is covered in "What Is GPT-5 in ChatGPT? Free Tier, Pricing, and Choosing a Model" and Gemini in "What Are Gemini's Latest Models? Pricing, Free Tier, and How to Use 3.5".
Summary
Claude's four models form a ladder: Fable 5 for long autonomous projects, Opus 4.8 for deep reasoning, Sonnet 5 for the balance of speed and intelligence, and Haiku 4.5 for the fastest, cheapest throughput.
For most work Sonnet 5 is the first choice. Escalate to Opus 4.8 when a tangled problem needs another level of accuracy, and push bulk routine work down to Haiku 4.5. That is the basic shape of a cost-versus-quality balance, with Fable 5 held in reserve for large jobs you want to hand over in one piece.
In the chat interface, the free plan covers Haiku and Sonnet. Reaching Opus 4.8 and Fable 5 means Pro ($20 a month, or an effective $17 billed annually) or above. On the API everything is pay-as-you-go, and Sonnet 5 carries introductory pricing of $2 input / $10 output through August 31, 2026. Layer the Batch API and prompt caching on top and the effective rate drops further still.
Trying Sonnet 5 on the free plan first, then moving to Pro if it falls short, is enough to get started. For the basics of the service itself, see "What Is Claude? Pricing, Choosing a Model, and Claude vs ChatGPT".
