How much does Claude Fable 5.1 cost per million tokens?

Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens on the standard first-party API. Cached reads (prompt cache hits) run $0.25 per million tokens — 97.5% off the standard input rate. Five-minute cache writes are $12.50 per million tokens and one-hour cache writes are $20 per million tokens. The standard rates on Fable 5.1 match GPT-6 Astra exactly on input, output, and 5-minute cache write, but the 0.025x cache read on Fable 5.1 is 4x cheaper than OpenAI's 0.1x (Anthropic pricing page, September 2026).

Is Claude Sonnet 5 pricing permanent?

Yes. Claude Sonnet 5 is $2 input / $10 output per million tokens as of September 2026, and Anthropic made the introductory $2/$10 permanent on August 10, 2026. The previously scheduled increase to $3/$15 on September 1, 2026 was cancelled. Sonnet 5 is now the default model for Free and Pro plans in Claude.ai, and the recommended mid-tier option on the API (Anthropic blog, August 2026).

What is the difference between Fable 5 and Fable 5.1?

Fable 5.1 is priced identically to Fable 5 on input ($10/MTok) and output ($50/MTok), but cuts the cache read rate from $1.00 to $0.25 per million tokens — a 75% discount on cache hits. Fable 5.1 and Mythos 5.1 both use the 0.025x cache multiplier, while every other Anthropic model uses the standard 0.1x. Cache writes ($12.50 5-minute, $20 1-hour) are unchanged. Fable 5.1 also bumps the knowledge cutoff to June 2026 (Anthropic pricing page, September 2026).

How much does Claude Opus 5 cost?

Claude Opus 5 is $5 input / $25 output per million tokens on the standard tier — half of Fable 5.1. Cache writes are $6.25 (5-minute) / $10 (1-hour) and cache reads are $0.50 per million tokens. Opus 4.8, 4.7, 4.6, and 4.5 all list at the same $5/$25 rates, making Opus 4.x the cheapest path to a 1M-context Claude model. Fast mode on Opus 5 and Opus 4.8 doubles the rate to $10 input / $50 output per million tokens (Anthropic pricing page, September 2026).

What does Claude Haiku 4.5 cost?

Claude Haiku 4.5 is $1 input / $5 output per million tokens on the standard tier — the cheapest Claude model in production as of September 2026. Cache writes are $1.25 (5-minute) / $2 (1-hour) and cache reads are $0.10 per million tokens. Haiku 4.5 supports a 200K context window and 64K max output — smaller than the 1M context on Fable, Opus, and Sonnet 5. Haiku 3.5 is retired on the first-party API but remains available on AWS Bedrock and Google Cloud at $0.80/$4 (Anthropic pricing page, September 2026).

How does Claude pricing differ on AWS Bedrock vs Google Vertex AI?

Claude pricing on AWS Bedrock and Google Vertex AI matches Anthropic's first-party rate exactly for the standard rates. Both platforms add a 10% regional premium on certain endpoints — for example, US inference on Anthropic adds a 1.1x token multiplier on supported models. US-only inference is available via direct Anthropic API and via Vertex AI in the US region; Bedrock supports US-only inference through cross-region inference profiles (Anthropic pricing notes, September 2026).

How much does Claude Batch API cost?

Claude Batch API requests are 50% off the standard rate across input, output, and cache writes. On Fable 5.1, Batch runs at $5 input / $25 output per million tokens with cache reads at $0.125 per million tokens. Batch returns results within 24 hours and is the cheapest path to a frontier Claude model. The 50% discount stacks with prompt caching, so a Fable 5.1 Batch workload with high cache reuse can run at under $0.13 per million tokens — 75x cheaper than the headline standard rate (Anthropic pricing page, September 2026).

Claude API Pricing 2026: Cost Per Million Tokens

Quick Answer

Claude API pricing 2026: Fable 5.1 $10 input / $50 output per million tokens with $0.25 cache reads (0.025x — 4x cheaper than the standard 0.1x); Sonnet 5 $2/$10 PERMANENT (intro pricing made permanent August 10, 2026); Opus 5 $5/$25; Haiku 4.5 $1/$5. Batch API 50% off; Bedrock and Vertex AI add 10% regional premium on certain endpoints.

Last verified: Sep 16, 2026.

At a glance

Claude Fable 5.1: $10 input / $50 output per 1M tokens; cache read $0.25 (0.025x)
Claude Mythos 5.1: $10/$50 same as Fable; Glasswing-restricted
Claude Sonnet 5: $2/$10 PERMANENT; default for Free and Pro plans
Claude Opus 5 / Opus 4.8: $5 input / $25 output per 1M tokens
Claude Haiku 4.5: $1 input / $5 output; cheapest in the family
Cache writes: $12.50 (5-min) / $20 (1-hour) for Fable 5.1
Batch API: 50% off standard; 24-hour turnaround
Regional uplift: 10% premium on Bedrock/Vertex regional endpoints; 1.1x US-only inference multiplier

Fable 5.1 cache reads are the cheapest in frontier AI

Claude Fable 5.1's $0.25-per-million-token cache read is 4x cheaper than OpenAI's cache read on GPT-6 Astra, and it is the single biggest cost lever for high-volume Claude deployments.

Anthropic uses two cache multipliers: 0.025x on Fable 5.1 and Mythos 5.1, and 0.1x on every other model including Opus 5, Sonnet 5, and Haiku 4.5. That means a Fable 5.1 cache hit is $0.25 per million tokens versus $1.00 on Fable 5, $0.50 on Opus 5, $0.20 on Sonnet 5, and $0.10 on Haiku 4.5. For a workload that reuses a 50,000-token system prompt across 10,000 requests per day, Fable 5.1 saves $375 per day versus Sonnet 5 on cache reads alone (Anthropic pricing page, September 2026).

The cache is automatic: any prompt prefix that has been seen in the last 5 minutes (or 1 hour if configured) is served from cache at the cache read rate. Cache writes happen once per prefix per TTL window, so the amortised cost depends on reuse rate. A 50,000-token prefix written once and reused 100 times in 5 minutes costs $0.0125 per request on Fable 5.1 — 800x cheaper than the headline input rate (eesel.ai, September 2026).

Sonnet 5 became the default in mid-2026

Claude Sonnet 5 became the default Claude model for Free and Pro plans in Claude.ai on June 30, 2026, and Anthropic made the introductory $2/$10 pricing permanent on August 10, 2026.

The launch pricing was originally set to expire August 31, 2026, with the standard rate scheduled to rise to $3 input / $15 output on September 1, 2026. Anthropic cancelled that increase in an August 10 edit to the Sonnet 5 launch post, citing Sonnet 5's competitive position and the cost benefit to power users (Anthropic blog, August 2026).

Sonnet 5 has a 1M-token context window and 128K max output — matching Fable 5.1 and Opus 5. For most workloads that do not need Fable 5.1's deeper reasoning, Sonnet 5 offers the best price-performance in the Claude family: $2/$10 is 5x cheaper than Fable 5.1 on input and 2.5x cheaper on Opus 5. Sonnet 4.6 and 4.5 remain available at $3/$15 for teams that prefer the older model family (Anthropic pricing page, September 2026).

Opus 5 vs Opus 4.x: same price, newer knowledge cutoff

Anthropic priced Opus 5 identically to Opus 4.8, 4.7, 4.6, and 4.5 — the choice is about capabilities, not cost.

All five Opus models list at $5 input / $25 output per million tokens. Cache writes are $6.25 (5-minute) / $10 (1-hour) and cache reads are $0.50 per million tokens on all five. Fast mode on Opus 5 and Opus 4.8 doubles the input and output rates to $10/$50 per million tokens. The Opus 4.x line is the cheapest path to a 1M-token Claude context window — Opus 5 does not have a price advantage over its predecessors (Anthropic pricing page, September 2026).

Opus 4.1 and Opus 4 are listed on Anthropic's pricing page but are retired on the first-party API. They remain available on AWS Bedrock and Google Cloud for legacy customers at the same $5/$25 rate. New Anthropic API customers cannot provision them. Migrating Opus 4.1 or 4.0 traffic to Opus 5 is a free path to a more recent knowledge cutoff (May 2026) at the same price (eesel.ai, September 2026).

What enterprise buyers should do next

Three actions for organisations evaluating Claude API pricing in 2026.

Pick Sonnet 5 unless you have a reason not to. Sonnet 5's $2/$10 permanent rate is the best price-performance in the Claude family. Fable 5.1 is justified for deep reasoning and long-horizon agentic work. Opus 5 is justified only when Sonnet 5's accuracy is insufficient. Haiku 4.5 is the cheapest Claude API option for high-volume classification.
Design for prompt caching from day one. Keep system prompts above 1,024 tokens and as stable across requests as possible. A 50K-token system prompt reused 100x in 5 minutes runs at $0.0125 per request on Fable 5.1 versus $0.50 per request without caching.
Negotiate on annual commitments. Anthropic offers volume discounts for enterprise contracts starting around 100M tokens/month. The pricing page lists standard rates only — committed-use discounts can cut 20-40% off headline rates for predictable workloads.

What to watch next

Three near-term datapoints. First, whether Mythos 5.1 remains Glasswing-restricted — Anthropic has not yet published when (or if) Mythos 5.1 general availability will lift. Second, the Claude 5.2 release window — Anthropic typically ships a model family refresh every 6-9 months; a Sonnet 5.5 or Fable 5.5 is plausible for Q1 2027. Third, the impact of the 1.1x US-only inference multiplier on competitive pricing for US-regulated workloads — this multiplier applies on Fable 5.1, Opus 5, and Sonnet 5, making Anthropic US-only endpoints 10% more expensive than global endpoints.