Create more. Sell more this Black Friday.Explore
Claude API Pricing 2026: Anthropic Cost Reference
guide

Claude API Pricing 2026: Anthropic Cost Reference

EvoLink Team
EvoLink Team
Product Team
April 16, 2026
Updated on October 21, 2026
14 min read
Updated October 21, 2026: Claude Opus 5.5 costs $4 input / $20 output per million tokens on Anthropic's standard API. Both token rates are 20% below Opus 4.7's $5 / $25; actual task cost also depends on token usage, caching, effort and retries. For current budgeting, use the official rates below. EvoLink route prices are listed separately later in this guide.
ModelInput / MTokOutput / MTokModel details
Claude Fable 5.1$10.00$50.00Fable 5.1
Claude Opus 5.5$4.00$20.00Opus 5.5
Claude Sonnet 5.5$2.00$10.00Sonnet 5.5
Claude Haiku 5.5 — prompt ≤100K tokens$0.10$0.50Haiku 5.5
Claude Haiku 5.5 — prompt >100K tokens$0.50$2.50Haiku 5.5 long-context pricing
All prices are USD per million tokens (MTok), verified against Anthropic's pricing documentation on October 21, 2026. These are standard input/output rates; caching, Batch API, fast mode, data residency and server-side tools have separate rates or modifiers. For Haiku 5.5, prompt length includes cache reads and writes. Above 100K prompt tokens, the higher tier applies to the whole request, including output, rather than only the excess input.
Anthropic's model guide recommends starting with Opus 5.5 for most workloads when you are unsure which model to choose. Fable 5.1 is the higher-capability option for demanding reasoning and long-horizon agent work when Opus 5.5 evaluations fall short. For a lower-cost route, compare Sonnet 5.5 with Opus 5.5 on your task.

TL;DR

Claude API pricing depends on which model you pick and which pricing scope you mean. This guide separates:

  • official Anthropic API pricing
  • long-context / caching / batch pricing rules
  • current EvoLink route pricing, where available
The following table preserves the April 16, 2026 snapshot for older-model cost comparisons. It is not the current model recommendation list; check Anthropic's model lifecycle documentation before planning a deployment on a legacy model.

As of this update, Opus 4.1 and Sonnet 4 have retired from the Claude API. Sonnet 4.5 is deprecated, with retirement scheduled for November 30, 2026. Historical prices do not establish current availability on Anthropic or an EvoLink route.

ModelInputOutputBest for
Claude Opus 4.7$5.00$25.00Flagship covered by the April 16 snapshot
Claude Opus 4.6$5.00$25.00Previous flagship for pinned deployments and migration comparisons
Claude Opus 4.5$5.00$25.00Previous-gen flagship, stable versioned ID
Claude Opus 4.1$15.00$75.00Agentic coding, multi-file refactors
Claude Sonnet 4.6$3.00$15.00Historical balanced-tier baseline
Claude Sonnet 4.5$3.00$15.00Earlier coding and agent deployments
Claude Sonnet 4.0$3.00$15.00Legacy deployments, historical 1M context beta
Claude Haiku 4.5$1.00$5.00Lower-cost option within the Claude 4 family

If you are comparing direct vendor pricing with gateway pricing, Anthropic's official rates are the baseline. EvoLink can expose lower route-specific pricing for the same Claude family, but that is a separate pricing scope from direct Anthropic billing.


Historical context: Anthropic released Claude Opus 4.7 on April 16, 2026 at the same headline $5 / $25 pricing as Opus 4.6. For that specific migration, read our Claude Opus 4.7 vs Claude Opus 4.6 guide. The July snapshot added Fable 5 at $10 / $50, Opus 4.8 at $5 / $25 and Sonnet 5 at $2 / $10; those older-model rates do not replace the current lineup above. See the dedicated Fable 5 pricing guide for that version.

Claude Opus 4.7 Pricing

Claude Opus 4.7 was the flagship covered by this guide's April 16 snapshot. For newer Claude models, use their dedicated model and pricing pages rather than treating this historical section as a live catalog.

ComponentRate (per MTok)
Input tokens$5.00
Output tokens$25.00
Cache write (5 min)$6.25
Cache write (1 hour)$10.00
Cache read$0.50 (0.1x input)
Batch input$2.50 (50% off)
Batch output$12.50 (50% off)
Context window: 1M tokens at standard Anthropic pricing Max output: 128K tokens Model ID: claude-opus-4-7

What changed from Opus 4.6?

The headline token rates are unchanged. Anthropic's current pricing documentation says the newer tokenizer used by Claude 4.7 and later models produces approximately 30% more tokens for the same text, with the actual increase depending on the content and workload. Equal token rates therefore do not guarantee equal task cost.

Claude Opus 4.6 Pricing

Claude Opus 4.6 is an earlier Opus flagship retained here for migration comparisons and existing pinned deployments. Opus 4.7 succeeded it in the April snapshot; neither is the current top-tier recommendation. Use the current model table above for a new deployment, then validate a migration against your own workload.
ComponentRate (per MTok)
Input tokens$5.00
Output tokens$25.00
Cache write (5 min)$6.25
Cache write (1 hour)$10.00
Cache read$0.50 (0.1x input)
Batch input$2.50 (50% off)
Batch output$12.50 (50% off)
Context window: 1M tokens at standard Anthropic pricing Max output: 128K tokens Model ID: claude-opus-4-6

When Opus 4.6 is still worth using

  • Existing deployments pinned to claude-opus-4-6
  • Migration baselines and regression testing
  • Cost comparisons against Opus 4.7 where tokenizer behavior matters
  • Teams that want to validate prompt behavior before moving to 4.7

Claude Opus 4.5 Pricing

Claude Opus 4.5 shares the same pricing as Opus 4.6 and is a stable previous-generation option with a pinned versioned ID.
ComponentRate (per MTok)
Input tokens$5.00
Output tokens$25.00
Cache write (5 min)$6.25
Cache read$0.50
Model ID: claude-opus-4-5-20251101

Claude Opus 4.1 Pricing

Claude Opus 4.1 carried higher token rates than Opus 4.5/4.6/4.7. The prices below are retained for historical billing comparisons; the model retired from the Claude API on August 5, 2026.

ComponentRate (per MTok)
Input tokens$15.00
Output tokens$75.00
Cache write (5 min)$18.75
Cache read$1.50
Model ID: claude-opus-4-1-20250805

Is Opus 4.1 still worth using?

Its historical input/output token rates were 3x those of Opus 4.6 or 4.7. Because Opus 4.1 has retired from the Claude API, it is no longer a direct-API deployment option; use these figures only to compare older bills or migration baselines.


Claude Sonnet 4.6 Pricing

Claude Sonnet 4.6 is the Sonnet-tier baseline in the April snapshot, with 128K max output and adaptive thinking; the older extended-thinking mode is deprecated. Its historical role does not make it the default choice over newer Sonnet models.
ComponentRate (per MTok)
Input tokens$3.00
Output tokens$15.00
Cache write (5 min)$3.75
Cache read$0.30
Batch input$1.50 (50% off)
Batch output$7.50 (50% off)
Context window: 1M tokens at standard Anthropic pricing Max output: 128K tokens Model ID: claude-sonnet-4-6

Claude Sonnet 4.5 Pricing

Claude Sonnet 4.5 is an earlier model used for coding and agent deployments. This section records its historical rates rather than a current default-model recommendation.
ComponentRate (per MTok)
Input tokens$3.00
Output tokens$15.00
Cache write (5 min)$3.75
Cache read$0.30
Context window: 200K tokens, with 1M long-context pricing still treated separately Model ID: claude-sonnet-4-5-20250929

Claude Sonnet 4.0 Pricing

Claude Sonnet 4.0 historically used separate long-context pricing above 200K tokens. It retired from the Claude API on June 15, 2026; the table below records its former rates.

ComponentRate (per MTok)
Input tokens$3.00
Output tokens$15.00
Input (>200K, 1M beta)$6.00
Output (>200K, 1M beta)$22.50
Cache write (5 min)$3.75
Cache read$0.30
Model ID: claude-sonnet-4-20250514

Claude Haiku 4.5 Pricing

Claude Haiku 4.5 is the fastest and cheapest Claude 4 model.
ComponentRate (per MTok)
Input tokens$1.00
Output tokens$5.00
Cache write (5 min)$1.25
Cache read$0.10
Batch input$0.50 (50% off)
Batch output$2.50 (50% off)
Context window: 200K tokens Model ID: claude-haiku-4-5-20251001

When to choose Haiku over Sonnet

  • Free-tier products where cost per request matters
  • Parallelized sub-agents and high-volume pipelines
  • Latency-sensitive chat and support agents
  • Tasks where your evaluations show Haiku 4.5 meets the required quality

Cost Comparison: Claude vs GPT vs Gemini

This comparison retains the model selection from April 2026, with base rates rechecked on October 21, 2026. These are older-model references, not a comparison of each provider's latest lineup. The listed base rates do not apply to every context length.

ModelInput/MTokOutput/MTokContextMax Output
Claude Opus 4.7$5.00$25.001M128K
Claude Opus 4.6$5.00$25.001M128K
Claude Sonnet 4.6$3.00$15.001M128K
Claude Haiku 4.5$1.00$5.00200K-
GPT-5.4$2.50$15.001.05M128K
GPT-5.2$1.75$14.00400K128K
Gemini 3.1 Pro Preview$2.00$12.001M64K
Pricing scope: GPT-5.4 costs $2.50/$15 per MTok at the base tier; prompts above 272K input tokens trigger 2x input and 1.5x output pricing for the full session. GPT-5.2 is $1.75/$14. Gemini 3.1 Pro Preview is $2/$12 for prompts up to 200K tokens and $4/$18 above that threshold. Context limits and pricing thresholds are different measures.
For a detailed migration breakdown, see Claude Opus 4.7 vs Claude Opus 4.6. For a broader model-by-model comparison, see GPT-5.4 vs Claude Opus 4.6 vs Gemini 3.1 Pro.

How to Reduce Claude API Costs

1. Use prompt caching

Prompt caching reuses repeated system prompts or context. On Anthropic's API, a 5-minute cache write costs 1.25x the base input rate and a 1-hour write costs 2x. Cache reads cost 0.05x on Opus 5.5 and Sonnet 5.5 ($0.20 and $0.10/MTok), 0.025x on Fable 5.1 ($0.25/MTok), and 0.1x on Haiku 5.5 and the older models listed here. These are reductions on cache-hit input, not on the whole request; writes and output still contribute to the bill. See official cache pricing.

2. Use batch processing

Batch API provides a 50% discount on both input and output tokens for asynchronous jobs. Ideal for nightly automation, data processing, and non-real-time workloads.

3. Pick the right model tier

Don't use Opus for tasks Sonnet can handle. Don't use Sonnet for tasks Haiku can handle. Match model capability to task complexity.

The following EvoLink public-page price snapshot was checked on October 21, 2026. Input and output prices are USD per MTok; each link opens the existing model pricing section.
EvoLink routeInput / MTokOutput / MTok
Claude Opus 5.5$3.800$19.000
Claude Fable 5.1$9.500$47.500
Claude Sonnet 5.5$1.900$9.500
Claude Haiku 5.5 — prompt ≤100K tokens$0.096$0.475
Claude Haiku 5.5 — prompt >100K tokens$0.478$2.375

Use your account's current route rates for an EvoLink budget. Check cache writes, cache reads, tool fees and minimum charges separately: lower input/output prices do not imply the same discount on every billing dimension. Anthropic's cache multipliers and Batch discounts should not be applied automatically to an EvoLink route. EvoLink's unified API lets you compare supported models using one API key and balance; choose a route by task quality and total cost.

5. Monitor and set budgets

Review usage and billing records by model, set an internal spending budget, and compare actual token consumption with your estimate before scaling traffic.


FAQ

How much does Claude API cost?

As of October 21, 2026, Anthropic's standard input/output rates per MTok are $4/$20 for Opus 5.5, $10/$50 for Fable 5.1 and $2/$10 for Sonnet 5.5. Haiku 5.5 costs $0.10/$0.50 for prompts up to 100K tokens and $0.50/$2.50 above that threshold. Caching, batch processing and other billing options change the effective cost.

What is the cheapest Claude API model?

Among the current models in Anthropic's standard price list, Claude Haiku 5.5 has the lowest input/output rates: $0.10/$0.50 per MTok for prompts up to 100K tokens. Above 100K prompt tokens, the entire request uses the higher $0.50/$2.50 tier. Include cached input when checking the threshold.

Is Opus 5.5 cheaper than Opus 4.7?

Yes, on standard input/output token rates: $4/$20 versus $5/$25 per MTok, a 20% reduction in both rates. That is not a guarantee of 20% lower task cost: token usage, effort, cache hits and retries can change when you migrate.

Is Claude API free?

Anthropic's standard API pricing is usage-based. Check your account for any trial, subscriber or promotional credits and their eligibility and expiry terms; do not assume free credits cover ongoing production usage.

How does Claude pricing compare to GPT?

The comparison above retains the April 2026 model selection, with base rates rechecked on October 21, 2026. For a new decision, compare the current models' input/output mix, cache rates, context tiers and quality on the same task. Keep direct vendor prices separate from EvoLink route prices; the older table does not establish which provider is cheapest today.

What is prompt caching and how much does it save?

Prompt caching reuses repeated context. Anthropic charges 1.25x the base input rate for 5-minute writes and 2x for 1-hour writes. Cache reads are 0.05x on Opus 5.5 and Sonnet 5.5, 0.025x on Fable 5.1, and 0.1x on Haiku 5.5 and the older models listed here. Savings apply to cache-hit input; the total bill also includes writes, output and any tools.

Does Claude have batch processing discounts?

Yes. Anthropic's Batch API provides a 50% discount on both input and output tokens. Batches have a 24-hour processing window; requests that do not finish in that window expire. See the Batch API documentation.

Where can I access Claude API at the best price?

Compare EvoLink's available Claude routes with Anthropic's direct prices using your actual token mix. The route table above is a dated input/output snapshot, not a guarantee of the lowest total bill. Confirm account-specific cache and tool rates before choosing.

Use EvoLink's unified API to access supported Claude models with one API key and balance. Start with the model that meets your task requirements, compare its current route rates and monitor usage before scaling traffic.


Current official rates and EvoLink public-page prices checked on October 21, 2026. The April model tables remain historical references. Verify the relevant provider or account rates before final budgeting.

Connect AI models through one API

Compare models and current pricing, then follow the API documentation to integrate your chosen model.