Claude API Pricing in 2026: Models, Costs, and Budget Examples
A current, worked guide to Claude API pricing: model rates, Sonnet 5 launch pricing, caching and batch discounts, tool charges, and realistic monthly budgets.
In this article

Quick answer: For standard API requests, Fable 5 is $10 input / $50 output per million tokens, Opus 4.8 is $5 / $25, Sonnet 5 is temporarily $2 / $10 through August 31, 2026 before moving to $3 / $15, and Haiku 4.5 is $1 / $5. All figures are USD.
Current Claude API model prices
Anthropic's current public model ladder gives teams four broad cost-and-capability points. Fable is the highest-capability widely released model. Opus targets complex agentic coding and enterprise work. Sonnet is the default balance of speed and intelligence. Haiku is the fastest and lowest-cost option.
- Claude Fable 5: $10 per million input tokens and $50 per million output tokens. 1M-token context and up to 128K synchronous output.
- Claude Opus 4.8: $5 input and $25 output per million tokens. 1M-token context and up to 128K synchronous output.
- Claude Sonnet 5 through August 31, 2026: introductory $2 input and $10 output per million tokens. On September 1, standard pricing becomes $3 input and $15 output. It has a 1M-token context window and up to 128K synchronous output.
- Claude Haiku 4.5: $1 input and $5 output per million tokens. 200K-token context and up to 64K synchronous output.
- Claude Mythos 5: the same $10 / $50 pricing and specifications as Fable 5, but limited to approved Project Glasswing customers rather than self-service access.
Those are base token rates. A request that uses 100,000 input tokens consumes 0.1 million input tokens. A response of 5,000 tokens consumes 0.005 million output tokens. Multiply each quantity by the relevant rate, then add the two parts.
The cost formula
For a standard request, cost equals input tokens divided by 1,000,000 times the model's input price, plus output tokens divided by 1,000,000 times the output price. Use actual usage fields from API responses for forecasting; word counts and character counts are only rough estimates.
Example request: 100,000 input tokens plus 5,000 output tokens costs about $1.25 on Fable 5, $0.625 on Opus 4.8, $0.25 on Sonnet 5 at the introductory rate, $0.375 on Sonnet 5 at the September standard rate, or $0.125 on Haiku 4.5.
Which model is the economical default?
The cheapest model per token is not automatically the lowest-cost model per accepted result. A model that needs repeated prompts, produces more reviewer corrections, or fails a task can cost more than a stronger model that finishes in one pass. Start with Sonnet 5 for most agentic coding, tool-use, support, extraction, and knowledge-work applications. Test Haiku when latency and volume dominate. Move to Opus or Fable only where measured quality, autonomy, or reviewer time justifies the premium.
- Use Haiku 4.5 for routing, tagging, simple extraction, short summaries, and other high-volume tasks after quality testing.
- Use Sonnet 5 as the first benchmark for general production work. Its temporary launch rate makes it especially attractive through August 31, but budget against the $3 / $15 standard rate for workloads that continue into September.
- Use Opus 4.8 for complex agentic coding and enterprise work that Sonnet cannot complete reliably enough.
- Use Fable 5 for the highest-capability long-running work after confirming its stricter safeguards and 30-day retention requirement fit the use case.
Worked budget example: document analysis
Suppose a workflow processes 1,000 document packages each month. Each package sends 25,000 input tokens and receives 2,000 output tokens. Monthly usage is 25 million input tokens and 2 million output tokens.
- Fable 5: 25 × $10 plus 2 × $50 = $350 per month.
- Opus 4.8: 25 × $5 plus 2 × $25 = $175 per month.
- Sonnet 5 introductory rate: 25 × $2 plus 2 × $10 = $70 per month.
- Sonnet 5 standard rate: 25 × $3 plus 2 × $15 = $105 per month.
- Haiku 4.5: 25 × $1 plus 2 × $5 = $35 per month.
The spread is meaningful, but the correct comparison is cost per approved analysis. If Haiku requires frequent escalation or misses important terms, route the harder documents to Sonnet or Opus instead of forcing one model across the entire workload.
Worked budget example: high-volume classification
A service classifies 100,000 short records per month. Each call averages 1,000 input tokens and 100 output tokens. That is 100 million input tokens and 10 million output tokens.
- Haiku 4.5: $100 input plus $50 output = $150 per month.
- Sonnet 5 introductory rate: $200 input plus $100 output = $300 per month.
- Sonnet 5 standard rate: $300 input plus $150 output = $450 per month.
- Opus 4.8: $500 input plus $250 output = $750 per month.
- Fable 5: $1,000 input plus $500 output = $1,500 per month.
This is the kind of workload where a tested Haiku-first route can halve or quarter token cost. Keep an escalation path for ambiguous records and measure the accepted-result rate separately for each route.
Prompt caching can change the economics
Prompt caching is valuable when the same long system prompt, document corpus, or conversation prefix appears across repeated calls. A five-minute cache write costs 1.25 times the base input rate, a one-hour write costs 2 times the base rate, and a cache read costs 0.1 times the base input rate. In other words, a cache hit is 90% cheaper than reprocessing the same input at the standard rate.
Caching only helps when calls actually reuse the cached prefix within its lifetime. Track cache creation and cache-read tokens independently. A theoretical 90% saving does not materialize if every request has a different prefix or misses the cache.
Batch processing is 50% cheaper
The Message Batches API discounts both input and output tokens by 50% for asynchronous work. That changes the document-analysis example at base rates from $70 to $35 on introductory Sonnet 5, or from $35 to $17.50 on Haiku. Batch is useful for backfills, nightly extraction, bulk classification, evaluation runs, and other work that does not need an immediate response.
Costs beyond model tokens
- Web search costs $10 per 1,000 searches in addition to the tokens used by search results and the model response. Failed searches are not billed.
- Web fetch has no separate request fee, but fetched content still contributes token usage.
- Tool definitions, tool calls, tool results, and the tool-use system prompt add tokens to the request.
- US-only inference on supported current models applies a 1.1-times multiplier. Partner cloud regions can have their own pricing.
- Opus 4.8 Fast Mode costs $10 input and $50 output per million tokens and cannot be combined with the Batch API.
- Newer Fable, Opus, and Sonnet models use an updated tokenizer. Anthropic says the same text can produce about 30% more tokens, depending on the workload, so remeasure rather than carrying forward old token estimates.
A practical forecasting process
- Capture input, cache-write, cache-read, output, and tool usage from real API responses for at least a representative pilot.
- Group usage by workflow and model, not only by application or customer.
- Calculate cost per accepted outcome and include reviewer time, retries, and fallbacks.
- Forecast Sonnet 5 at its September $3 / $15 rate unless the workload ends before August 31.
- Add separate assumptions for batch share, cache-hit rate, web searches, data residency, and cloud-provider premiums.
- Set budget and anomaly alerts, then review the largest token consumers weekly until the pattern stabilizes.
A durable model router usually beats a one-model mandate. Use the least expensive model that meets the quality bar, route exceptions upward, and keep the evidence needed to revisit that decision when pricing or model behavior changes.
Primary sources and update policy
Pricing and specifications were checked July 21, 2026 against Anthropic's official Claude Platform pricing page and model overview, plus the Sonnet 5 launch announcement. Prices, promotional periods, regional multipliers, and model availability can change. Verify the official pages before a procurement decision.
Official pricing: https://platform.claude.com/docs/en/about-claude/pricing Model overview: https://platform.claude.com/docs/en/about-claude/models/overview Sonnet 5 launch pricing: https://www.anthropic.com/news/claude-sonnet-5
Get the Claude playbook in your inbox.
One weekly email for Claude and Claude Code users. Real workflows, no hype. Subscribe and we send you The Claude Power-User Cheatsheet.
— ¶ —

Luke Thompson
Luke Thompson is the founder of The Operations Guide, LLC and editor of The Claude Insider. Based in Jonesborough, Tennessee, he has spent years building AI-augmented business systems and automation workflows for operators and teams. He began working with large language models in production well before the current wave of consumer AI tools, integrating them into client workflows, content pipelines, and operational infrastructure. At The Claude Insider, he writes about Claude with the specificity of someone who uses it daily as a professional tool — not as a reviewer or commentator, but as a builder. His coverage focuses on what actually works: prompt patterns, API integration strategies, agentic workflows, and the real-world tradeoffs that practitioners face. He is not affiliated with Anthropic, PBC.
Articles are researched and drafted with AI assistance, reviewed and edited by Luke Thompson.
Know where AI can pay off in your company.
Take the free two-minute AI Readiness Assessment. See your score, the two gaps holding you back, and the next move worth making.


