Claude Fable 5 API Cost Calculator

Official Anthropic Pricing Snapshot

According to Anthropic pricing docs, Claude Fable 5 uses the following rates for Anthropic Direct API: last checked 2026-06-12. Provider marketplace rates may vary.

Prices shown are for Anthropic Direct API as of 2026-06-12. Provider marketplace availability and pricing should be verified in the relevant provider console. Prices can change; always check the official Anthropic pricing page before making cost decisions.

Token Cost Formula

Estimate Claude Fable 5 API cost using the base formula:

For example:

• 10K input tokens, 5K output tokens: (10,000 × $10 / 1,000,000) + (5,000 × $50 / 1,000,000) = $0.10 + $0.25 = $0.35 per request
• 200K input tokens, 50K output tokens: (200,000 × 0.00001) + (50,000 × 0.00005) = $4.50 per request
• 1M input tokens, 100K output tokens: (1,000,000 × 0.00001) + (100,000 × 0.00005) = $15.00 per request

Cache Pricing and Batch Pricing

Claude Fable 5 supports cached context and batch processing, which reduce cost significantly when used appropriately.

Prompt Caching

When you include a cached context using a previous cache write:

• Cache writes are billed at $12.50/MTok (5-min TTL) or $20.00/MTok (1-hour TTL) — charged once when the cache is created
• Cache hits are billed at $1.00/MTok — significantly cheaper than full input pricing
• Cache refreshes are also billed at $1.00/MTok

For example, a 50K-token cache write with 1-hour TTL costs:

Each subsequent request using that cache hit costs $1.00/MTok for the 50K cache plus $50/MTok for the new output tokens.

Batch Processing

Batch API pricing is half the price of standard API pricing:

• Batch input: $5.00/MTok (50% of base input)
• Batch output: $25.00/MTok (50% of base output)

Batch is suitable for non-real-time workloads. Turnaround time is longer than synchronous API calls.

Context Window Impact

Claude Fable 5's 1M token context window means high-context tasks accumulate significant input costs. For coding agents, each file read, conversation turn and tool call result adds to the token count.

Context window cost factors:

• Large file reads — hundreds to thousands of tokens per file
• Conversation history — grows with every turn, all counted as input
• Tool call results — included in context and billed as input tokens
• System prompts — counted toward the input token budget

For long agentic sessions, monitor tokens_in and tokens_out from your request logs. A session with 10 tool calls against a 100K-token context can easily reach 1M+ input tokens, resulting in $10+ per session at standard rates.

Tool Use and Agent Overhead

Using Claude Fable 5 with tools in an agentic workflow adds token overhead beyond a simple chat:

• Tool definitions and descriptions count toward input tokens
• Tool call results (file contents, command output, search results) are appended to context as input
• Each tool call round involves reasoning tokens plus the tool output tokens
• Multiple parallel or sequential tool calls multiply the overhead

In a coding agent scenario, a session might include:

• Initial system prompt and user task (~2K tokens input)
• Project context loading (~20K tokens input)
• Three file reads (~15K tokens input)
• Three tool call round-trips (~30K tokens input, ~5K tokens output)
• Final response (~2K tokens output)

Estimated cost:

With prompt caching, if the project context was previously cached, the input cost for the second session drops significantly. Cache hits cost $1.00/MTok vs. $10.00/MTok for uncached input.

Failed Request / Retry Billing Notes

Failed requests and retries can affect your Claude Fable 5 bill depending on how your provider handles them:

• Timeout errors — if the provider bills on request initiation, the input tokens sent before timeout may still be charged
• Rate limit retries — each retry sends the full context again, compounding input token cost
• Validation errors — usually not billed, but check your provider's policy
• Partial output — if a request is interrupted mid-stream, output tokens generated before interruption are typically billed

Small Prepaid Test Checklist

Before scaling Claude Fable 5 usage, run a small prepaid test to verify actual cost behavior:

• ✓Send one minimal request and record tokens_in, tokens_out and the response time
• ✓Compare estimated cost (tokens × rate) against your provider dashboard deduction
• ✓Test with prompt caching to measure cache write vs. cache hit cost difference
• ✓Test one agentic tool call round-trip and measure the token increment
• ✓Simulate a timeout error and check whether tokens were billed before timeout
• ✓Review request IDs and usage logs for at least 5 requests to establish a cost baseline

Example Cost Scenarios

Simple chat (no tools, no cache)

1K input tokens, 500 output tokens:

Coding agent session with file reads

80K input tokens (context + file reads), 10K output tokens:

Coding agent with cached context (1-hour TTL)

50K cache write (one-time) + 50K cache hit + 10K output:

Batch processing (non-real-time)

200K input tokens, 50K output tokens via Batch API:

vs. Standard API for the same tokens: $2.00 + $2.50 = $4.50 (batch saves 50%).

Related Guides

Frequently Asked Questions