Claude models compared: Fable, Opus, Sonnet, Haiku

Claude model comparison: the four current models side by side

Claude is a family of AI models made by Anthropic. The name tells you the tier and the number the generation, which is why Haiku is still at 4.5 while the other three are at 5 or later. "What Is Claude AI? Free Plan, Models, vs ChatGPT" covers what the Claude service itself offers.

Anthropic's comparison table lines up the four like this. All of them read text and images, use tools and handle multiple languages, so they differ in depth, speed and cost, not in missing features.

Fable 5.1 and Opus 5.5 are built for long-running agentic tasks

The top two are the ones to give a whole job to. Anthropic pitches Fable 5.1 at demanding reasoning and long-horizon agentic work and Opus 5.5 at long-running agentic coding and knowledge work. Both run adaptive thinking, where the model decides how long to reason before each answer, and on these two you cannot switch it off.

All that reasoning also makes them the slowest pair. Fable 5.1 plans and checks its own work, and Anthropic rates its speed "Slower", two steps below Sonnet 5.5's "Fast". Opus 5.5 is rated "Moderate", and its effort setting (how hard the model thinks by default) starts at medium rather than high.

Claude Sonnet 5.5 and Haiku 4.5 trade depth for speed

Anthropic's Academy calls Sonnet the daily driver, fast enough for back-and-forth chat and capable enough for most coding, writing and analysis. Sonnet 5.5 also has adaptive thinking, but unlike on the top two, you can turn it off.

Haiku 4.5 is the lightest and fastest of the four. It suits short answers, basic summaries and field extraction, and in the Claude app it eats the least of your usage limit. Instead of adaptive thinking it has extended thinking, which you switch on yourself when a question needs step-by-step reasoning.

Haiku 4.5 has less context and an earlier knowledge cutoff

The context window is the amount of text a model can take in one request, your files and the conversation so far included. Fable 5.1, Opus 5.5 and Sonnet 5.5 all take 1M tokens, while Haiku 4.5 stops at 200K. On Haiku, a long contract or a large codebase may not fit in a single request.

Haiku 4.5 is also the least up to date. Its reliable knowledge stops in February 2025, well before the other three (January to June 2026). Anthropic lists its retirement as no sooner than October 15, 2026, and a successor has already been announced (see Claude Haiku 5.5 API pricing and benchmarks).

Claude pricing by model, from $1 to $10 per million input tokens

Claude API input price per million tokens (USD)

Each step up doubles the rate until Fable 5.1, which costs 2.5 times Opus 5.5. The same job on Fable 5.1 costs ten times what it costs on Haiku 4.5.

$02.557.510

$10Fable 5.1

$4Opus 5.5

$2Sonnet 5.5

$1Haiku 4.5

There are two ways to pay for Claude. The API charges for every token your software sends and receives, and a claude.ai plan charges a flat monthly fee for the chat app, up to a usage limit. API pricing comes first below, then the plans.

Claude API pricing: output costs five times input on every model

API prices are per million tokens (MTok), split into the input you send and the output the model writes. On all four current models, output costs exactly five times input.

For a job that sends 1 million input tokens and gets 200,000 output tokens back, the bill is $2 on Haiku 4.5, $4 on Sonnet 5.5, $8 on Opus 5.5 and $20 on Fable 5.1. At the same volume, Fable costs ten times what Haiku does.

The Claude Batch API and prompt caching cut the per-token rate

Jobs that can wait, like overnight tagging or bulk summaries, can run through the Batch API at half price on both input and output. That takes Opus 5.5 down to $2 input and $10 output, the same as standard Sonnet 5.5.

Prompt caching pays off when you send the same long instructions or documents again and again. On Sonnet 5.5 and Haiku 4.5 a cache read costs a tenth of the input price, while the top two get a bigger cut, 2.5% of input on Fable 5.1 ($0.25) and 5% on Opus 5.5 ($0.20). The two discounts stack, which matters most for agents that reread the same context at every step.

Claude Free stops at Sonnet, Pro adds Opus, Max includes Fable

Which models show up in the claude.ai model picker depends on your plan. Anthropic's current pricing page gives Fable a row of its own, and that row treats Pro and Max differently.

The Pro subscription does not include Fable; Pro users pay for it with usage credits at standard API rates, so choose Max if you want Fable for everyday work. Anthropic's Academy tutorial says Pro and Max add Opus and Fable, but it still names Opus 5. Go by the pricing page, which already lists Opus 5.5.

Which Claude model to use, and when to switch

No Claude model is best for everything, but Anthropic does name a starting point, and it depends on where you use Claude. The chat-focused Academy points to Sonnet and the developer docs to Opus 5.5. Either way the rule is the same: go down a tier when the task is simple and frequent, and go up only when the answer is not good enough.

In the Claude app, start with Sonnet 5.5

claude.ai users who are not sure which model to pick should start with Sonnet, Anthropic's Academy says. Sonnet covers writing, analysis and most coding, is on every plan including Free, and uses a moderate share of your usage limit.

If an answer feels too shallow, try the Effort setting in the model picker before changing models. Turning it up makes Claude think longer on the same model, and turning it down gives quicker replies that use less of your limit.

On the Claude API, Anthropic recommends starting with Opus 5.5

For most workloads, Anthropic's developer docs recommend Opus 5.5 rather than Sonnet. At launch, Anthropic said Opus 5.5 matches Fable 5.1 on most work for less than half Fable's per-token price. "Claude Opus 5.5 pricing, benchmarks, API changes" covers that release in detail.

Sonnet 5.5 still makes sense on the API when speed or cost per call matters more than depth, as in a chat widget. For an agent that must plan, write code and fix its own mistakes, the docs advise starting on Opus 5.5 and measuring before paying for Fable.

Use Claude Haiku 4.5 for quick, high-volume jobs

Haiku 4.5 is worth it when the same simple step runs thousands of times. Sorting support tickets, pulling fields from forms or writing one-line summaries rarely needs a bigger model, and at $1 input and $5 output it costs half as much as Sonnet 5.5.

A common pattern is to let Haiku 4.5 take the first pass and send only the unclear cases up to Sonnet 5.5 or Opus 5.5. Check the context window first, though, since anything over 200K tokens has to go to one of the other three.

Move up to Claude Fable 5.1 when Opus 5.5 falls short

Fable 5.1 is for work Opus 5.5 has already tried and not managed: very long builds, dense technical material, or jobs you want to describe once and leave running. It costs 2.5 times Opus 5.5 per token and answers more slowly, so it is a poor default.

The Claude app adds one catch. On biology and security questions, Claude answers with Opus even when Fable is selected, so there is no reason to start those conversations on Fable.

Older Claude models still on the API, and their retirement dates

Older model names still appear in tutorials, tools and existing code. Many still work on the API, but none costs less per token than the current model that replaces it, so moving to the current model never raises your rate.

Claude Opus 5, Fable 5, Sonnet 5 and Opus 4.5–4.8 still work

Opus 5 is still available at $5 input and $25 output, which is more than Opus 5.5 ($4 and $20). Opus 4.8, 4.7, 4.6 and 4.5 cost the same $5 and $25, and Fable 5 is priced like Fable 5.1 ($10 and $50). Sonnet 5 stays available until at least June 30, 2027 at Sonnet 5.5 prices, cache and batch rates included; Sonnet 4.6 and 4.5 cost $3 input and $15 output, more than Sonnet 5.5 ($2 and $10).

The exception is Opus 4.1, which was retired on August 5, 2026. Calls to a retired model fail instead of falling back to a newer one, so code still using it needs a new model ID.

Sonnet 4.5 and Haiku 4.5 are next in line for retirement

Sonnet 4.5 (not before September 29, 2026) and Haiku 4.5 (not before October 15, 2026) head Anthropic's retirement list, followed by Opus 4.5 (not before November 24, 2026). These are the earliest possible dates, and Anthropic gives at least 60 days' notice before retiring a public model.

The dates cover the platforms Anthropic runs itself: the Claude API, Microsoft Foundry and Claude Platform on AWS. If you use Claude through Amazon Bedrock or Google Cloud, check that provider's retirement table instead.

Summary: start on Sonnet in the app and Opus on the API

Claude's four current models run from fast and cheap to slow and deep: Haiku 4.5 for speed and volume, Sonnet 5.5 for everyday work, Opus 5.5 for long coding and knowledge tasks, and Fable 5.1 for the hardest long-running jobs. API input prices rise from $1 to $10 per million tokens, and output always costs five times input.

In the Claude app, start on Sonnet 5.5, which comes with the Free plan; Opus needs Pro, and Fable comes with Max or through usage credits on Pro. On the API, start on Opus 5.5 and move to Fable 5.1 only when your tests show Opus falling short.