Claude Fable 5
Long-running agents and the hardest work
Anthropic's most capable widely released model for complex reasoning, science, math, and advanced coding.
Context / output: 1M / 128K
Anthropic model guide
Claude Models, Context, and Token Cost Calculator
Plan Claude workloads without treating every model as interchangeable. Compare Fable, Sonnet, Opus, and Haiku, then estimate the cost of long documents, coding sessions, support analysis, or agents at your expected traffic.
A useful Claude model shortlist
Anthropic frames model selection as a tradeoff among capability, speed, and cost. This shortlist keeps that decision concrete. Begin with the least expensive model that might pass your evaluation, then move up only when the result is measurably better for the task.
Model
Where it fits
Context / output
Claude Sonnet 5
Strong capability with lower latency and cost
The balanced choice for production agents, coding, analysis, and general workloads that still need substantial intelligence.
Context / output: 1M / 128K
Claude Opus 5
Complex agentic coding and enterprise work
Anthropic's latest Opus model for deep reasoning, agentic coding, long-horizon tasks, and demanding enterprise workflows.
Context / output: 1M / 128K
Claude Haiku 4.5
Fast, high-volume requests
The cost-effective choice for classification, routing, short summaries, and latency-sensitive product interactions.
Context / output: 200K / 64K
Interactive cost estimate
Estimate Anthropic API cost
Set the size of a Claude request and its monthly volume. The table compares the Anthropic models in the site's current pricing data.
Request pattern
Example workloads
Highest monthly estimate
$1,360
Claude Fable 5
Lowest monthly estimate
$136.0000
Claude Haiku 4.5
Estimated standard text cost
Rates are per 1M tokens. Results are sorted from highest to lowest monthly estimate.
Claude Fable 5
Context 1M / 128K
$1,360
per month
Input
$10
Cache
$1
Output
$50
$0.6800 per request
Claude Mythos 5 (limited availability)
Context 1M / 128K
$1,360
per month
Input
$10
Cache
$1
Output
$50
$0.6800 per request
Claude Opus 4.8
Context 1M / 128K
$680.0000
per month
Input
$5
Cache
$0.5
Output
$25
$0.3400 per request
Claude Opus 5
Context 1M / 128K
$680.0000
per month
Input
$5
Cache
$0.5
Output
$25
$0.3400 per request
Claude Sonnet 4.6
Context 1M / 128K
$408.0000
per month
Input
$3
Cache
$0.3
Output
$15
$0.2040 per request
Claude Sonnet 5
Context 1M / 128K
$272.0000
per month
Input
$2
Cache
$0.2
Output
$10
$0.1360 per request
Claude Haiku 4.5
Context 200K / 64K
$136.0000
per month
Input
$1
Cache
$0.1
Output
$5
$0.0680 per request
| Model | Rates per 1M | Per request | Per month |
|---|---|---|---|
Claude Fable 5 Context 1M / 128K | Input $10 Cache read $1 Output $50 | $0.6800 | $1,360 |
Claude Mythos 5 (limited availability) Context 1M / 128K | Input $10 Cache read $1 Output $50 | $0.6800 | $1,360 |
Claude Opus 4.8 Context 1M / 128K | Input $5 Cache read $0.5 Output $25 | $0.3400 | $680.0000 |
Claude Opus 5 Context 1M / 128K | Input $5 Cache read $0.5 Output $25 | $0.3400 | $680.0000 |
Claude Sonnet 4.6 Context 1M / 128K | Input $3 Cache read $0.3 Output $15 | $0.2040 | $408.0000 |
Claude Sonnet 5 Context 1M / 128K | Input $2 Cache read $0.2 Output $10 | $0.1360 | $272.0000 |
Claude Haiku 4.5 Context 200K / 64K | Input $1 Cache read $0.1 Output $5 | $0.0680 | $136.0000 |
This is a planning estimate for text requests. Prompt-cache writes, Batch API discounts, tools, regional pricing, and other separately billed features are not included.
How to size a Claude request
Long-context work tends to hide cost in places a word count misses. A document review request may include a system prompt, the source document, a rubric, prior turns, tool definitions, and a long answer. Measure each piece before deciding whether the whole document belongs in every call.
Long document review
Keep source text separate from the expected answer in your estimate. If the document rarely changes, evaluate prompt caching instead of resending it at the ordinary input rate forever.
Coding sessions
Repository context, tool results, patches, and conversation history accumulate quickly. Summarize stale turns and send only the files the next step genuinely needs.
Support analysis
A short classification or summary often fits Haiku. Escalate ambiguous or high-impact cases after the first pass rather than paying a top-tier rate for every ticket.
Long-running agents
Budget for repeated turns, not just the opening prompt. An agent that loops through tools can multiply both input context and output much faster than a single chat request.
Three ways to control Claude cost
Route by difficulty
Send routine classification and extraction to Haiku, keep Sonnet as the general default, and reserve Fable for requests that prove they need it.
Reuse stable context
Prompt caching can help when the same large instructions or documents appear across requests. Include write and storage rules in the final business case.
Batch work that can wait
Anthropic offers Batch API pricing for asynchronous jobs. It can change the economics of evaluations, classification runs, and offline document processing.
Claude token calculator FAQ
Why can Claude token counts differ between model generations?
Anthropic documents model-specific tokenizer changes. Newer Claude families can produce a different token count for the same text, so the official counting endpoint for your selected model is the best check before enforcing a hard limit.
Which Claude model is the best default?
Sonnet is usually the sensible model to evaluate first when you need a balance of capability, speed, and cost. Haiku fits simpler high-volume work, while Fable is aimed at the hardest jobs and long-running agents.
How should I estimate a long-document Claude workflow?
Count the instructions, document text, conversation history, and expected answer separately. Then multiply the full request estimate by its monthly volume and account for any cache writes or Batch API usage outside this calculator.
Does prompt caching make a long prompt free?
No. Eligible cache reads can be cheaper, but cache writes and storage rules still matter. Dynamic parts of the request continue to use ordinary input pricing.