Claude Opus API pricing by version, from Opus 4 to Opus 5 and the Fable tier above it
Every Claude Opus version from 4 to 5 on one price ladder, with the Fable and Mythos tiers above it, batch and fast mode rates, and three agent sessions priced across the line.
Every figure captured 6 September 2026 from the official Anthropic pricing page and two public price registries. Prices change, so read this as a dated snapshot.
$5 per million input tokens and $25 per million output tokens. That is the Claude Opus API price for every Opus version from 4.5 onward (4.5, 4.6, 4.7, 4.8 and Opus 5) on the official Anthropic pricing page as of 6 September 2026. Opus 4 and Opus 4.1 are listed at $15 and $75, three times as much, and both carry a retired label. Above Opus, Claude Fable 5 and Fable 5.1 are listed at $10 input and $50 output, so the top of the ladder is priced at two thirds of what Opus 4.1 used to cost, not at the old $15/$75.
The cache columns scale with base input on every row except two. Cache hits are 0.1x base input on all Opus versions ($0.50 on Opus 5, $1.50 on Opus 4.1), and 0.025x on Fable 5.1 and Mythos 5.1 ($0.25), which the official footnote calls out separately.
Price ladder by version, USD per 1M tokens
Source, docs.anthropic.com pricing page, model pricing table, captured 6 September 2026. Context window column from the LiteLLM registry, same day.
| Model | Base input | 5m cache write | 1h cache write | Cache hit | Output | Context (LiteLLM) |
|---|---|---|---|---|---|---|
| Claude Fable 5.1 | $10 | $12.50 | $20 | $0.25 | $50 | 1,000,000 |
| Claude Mythos 5.1 (limited availability) | $10 | $12.50 | $20 | $0.25 | $50 | 1,000,000 |
| Claude Fable 5 | $10 | $12.50 | $20 | $1 | $50 | 1,000,000 |
| Claude Mythos 5 (limited availability) | $10 | $12.50 | $20 | $1 | $50 | 1,000,000 |
| Claude Opus 5 | $5 | $6.25 | $10 | $0.50 | $25 | 1,000,000 |
| Claude Opus 4.8 | $5 | $6.25 | $10 | $0.50 | $25 | 1,000,000 |
| Claude Opus 4.7 | $5 | $6.25 | $10 | $0.50 | $25 | 1,000,000 |
| Claude Opus 4.6 | $5 | $6.25 | $10 | $0.50 | $25 | 1,000,000 |
| Claude Opus 4.5 | $5 | $6.25 | $10 | $0.50 | $25 | 200,000 |
| Claude Opus 4.1 (retired, except on Bedrock and Google Cloud) | $15 | $18.75 | $30 | $1.50 | $75 | 200,000 |
| Claude Opus 4 (retired, except on Google Cloud) | $15 | $18.75 | $30 | $1.50 | $75 | 200,000 |
| Claude Sonnet 5 (for comparison) | $2 | $2.50 | $4 | $0.20 | $10 | 1,000,000 |
The context column is the LiteLLM max_input_tokens value. LiteLLM shows 200,000 for claude-opus-4-20250514, claude-opus-4-1 and claude-opus-4-5, and 1,000,000 for claude-opus-4-6 and every later key, and OpenRouter reports the same figures. The official page agrees in prose, saying that Claude 4.6 and later models include the full 1M token context window at standard pricing, and that a 900k-token request is billed at the same per-token rate as a 9k-token request.
One thing the per-token rate hides. The official page says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text. A 4.7, 4.8 or Opus 5 request on identical text bills more tokens than the same request on 4.6 at the same rate. The worked examples below hold token counts fixed.
Batch API rates
The official page describes the Batch API as asynchronous processing of large volumes of requests with a 50% discount on both input and output tokens. Every row below is exactly half the standard row above.
Source, docs.anthropic.com pricing page, batch processing table, captured 6 September 2026. USD per 1M tokens.
| Model | Batch input | Batch output |
|---|---|---|
| Claude Fable 5.1 | $5 | $25 |
| Claude Fable 5 | $5 | $25 |
| Claude Opus 5 | $2.50 | $12.50 |
| Claude Opus 4.8 | $2.50 | $12.50 |
| Claude Opus 4.7 | $2.50 | $12.50 |
| Claude Opus 4.6 | $2.50 | $12.50 |
| Claude Opus 4.5 | $2.50 | $12.50 |
| Claude Opus 4.1 (retired, except on Bedrock and Google Cloud) | $7.50 | $37.50 |
| Claude Opus 4 (retired, except on Google Cloud) | $7.50 | $37.50 |
| Claude Sonnet 5 (for comparison) | $1 | $5 |
The Mythos 5 and 5.1 batch rows are also $5 and $25. Batch API and prompt caching discounts can be combined, and fast mode is not available with the Batch API.
Fast mode, Opus 5 and Opus 4.8 only
The official page says that fast mode, in research preview, provides significantly faster output for Claude Opus 5 and Claude Opus 4.8 at premium pricing, that fast mode pricing applies across the full context window, including requests over 200k input tokens, and that fast mode is available on the Claude API (first-party) only. The premium is exactly 2x the standard rate.
Source, docs.anthropic.com pricing page, fast mode pricing table, captured 6 September 2026. USD per 1M tokens.
| Model | Input | Output |
|---|---|---|
| Claude Opus 5 / Claude Opus 4.8 | $10 | $50 |
For the two versions just below, the page is explicit. Fast mode is not available on Claude Opus 4.7 (requests with speed: "fast" return an error) or Claude Opus 4.6 (requests run at standard speed and are billed at standard rates). So a 4.7 caller gets an error and a 4.6 caller silently gets normal speed at the normal $5/$25. Prompt caching multipliers and data residency multipliers apply on top of fast mode pricing.
Opus versus Sonnet, the direct ratio
Ratio of official rates, Opus 5 divided by Sonnet 5 and by Sonnet 4.6, computed from the 6 September 2026 table.
| Pair | Base input | Output | Cache hit | Batch input / output |
|---|---|---|---|---|
| Opus 5 ($5/$25) vs Sonnet 5 ($2/$10) | 2.5x | 2.5x | 2.5x ($0.50 vs $0.20) | 2.5x ($2.50/$12.50 vs $1/$5) |
| Opus 5 ($5/$25) vs Sonnet 4.6 ($3/$15) | 1.67x | 1.67x | 1.67x ($0.50 vs $0.30) | 1.67x ($2.50/$12.50 vs $1.50/$7.50) |
| Opus 4.1 ($15/$75) vs Sonnet 5 ($2/$10) | 7.5x | 7.5x | 7.5x ($1.50 vs $0.20) | 7.5x ($7.50/$37.50 vs $1/$5) |
The official page notes that the Sonnet 5 $2/$10 pricing, announced at launch as introductory through August 31, 2026, is now the standard price, and the scheduled increase to $3/$15 on September 1, 2026 will not occur. So the 2.5x gap is against Sonnet 5's standard price, not a promotional one.
Three worked examples
All three use one coding agent session of 150,000 input tokens and 12,000 output tokens, computed in Python from the rates above. Opus 4.1 is the old tier, Opus 5 the current one, Sonnet 5 the tier below and Fable 5.1 the tier above.
Example 1, one session at the standard rate with no caching. 150,000 input, 12,000 output.
| Model | Input cost | Output cost | Session total |
|---|---|---|---|
| Opus 4.1 | $2.25 | $0.90 | $3.15 |
| Opus 5 | $0.75 | $0.30 | $1.05 |
| Sonnet 5 | $0.30 | $0.12 | $0.42 |
| Fable 5.1 | $1.50 | $0.60 | $2.10 |
Example 2 is the same session with 120,000 of the input tokens served as cache hits and 30,000 at the base input rate, the shape of an agent loop where the system prompt, tools and repository context are stable and only the latest turn is new.
Example 2, 120,000 cache hit tokens, 30,000 fresh input, 12,000 output. The last column is the one-time 5 minute cache write for those 120,000 tokens.
| Model | Cache hits | Fresh input | Output | Session total | Saving vs example 1 | 5m cache write, once |
|---|---|---|---|---|---|---|
| Opus 4.1 | $0.18 | $0.45 | $0.90 | $1.53 | $1.62 (51.4%) | $2.25 |
| Opus 5 | $0.06 | $0.15 | $0.30 | $0.51 | $0.54 (51.4%) | $0.75 |
| Sonnet 5 | $0.02 | $0.06 | $0.12 | $0.20 | $0.22 (51.4%) | $0.30 |
| Fable 5.1 | $0.03 | $0.30 | $0.60 | $0.93 | $1.17 (55.7%) | $1.50 |
Fable 5.1 saves a larger share because its cache hit rate is 0.025x rather than 0.1x. With that much of the prompt cached, a Fable 5.1 session ($0.93) costs 1.82x a cached Opus 5 session ($0.51), down from the 2x gap between the uncached sessions ($2.10 against $1.05). The official page says caching pays off after one cache read for the 5 minute duration, and after two reads for the 1 hour duration. The prompt caching break-even calculator runs that arithmetic for any token count.
Example 3, 200 sessions in a month on the Batch API rate, no caching. 30,000,000 input and 2,400,000 output tokens in total.
| Model | Batch input | Batch output | Month total | Standard rate would be |
|---|---|---|---|---|
| Opus 4.1 | $225.00 | $90.00 | $315.00 | $630.00 |
| Opus 5 | $75.00 | $30.00 | $105.00 | $210.00 |
| Sonnet 5 | $30.00 | $12.00 | $42.00 | $84.00 |
| Fable 5.1 | $150.00 | $60.00 | $210.00 | $420.00 |
An interactive agent cannot wait for a batch, so read example 3 as the cost of the same token volume done as offline work, nightly code review for instance. At batch rates a month of Opus 5 ($105) is a third of Opus 4.1 ($315) and half of Fable 5.1 ($210).
What changed across the versions
The table shows exactly one price break inside the Opus line. Opus 4 (LiteLLM key claude-opus-4-20250514) and Opus 4.1 are $15 input, $18.75 for a 5 minute cache write, $30 for a 1 hour write, $1.50 per cache hit and $75 output. Opus 4.5 dropped every one of those numbers to one third, $5, $6.25, $10, $0.50 and $25, and 4.6, 4.7, 4.8 and Opus 5 kept them unchanged. The batch rows moved in lockstep, from $7.50/$37.50 to $2.50/$12.50.
There is still a tier above Opus, just not at the old number. Fable 5 and Fable 5.1 sit above Opus at $10/$50, which is 2x Opus 5 and two thirds of Opus 4.1. The one difference between the two Fable rows is the cache hit rate, $1 on Fable 5 and $0.25 on Fable 5.1. Mythos 5 and 5.1, both marked limited availability, mirror the Fable rows exactly.
Two non-price changes matter for a bill. The context window went from 200K on Opus 4.5 and earlier to 1M on 4.6 and later with no surcharge, and the tokenizer changed at 4.7, adding roughly 30% more tokens for the same text. Fast mode is a 2x premium on Opus 4.8 and Opus 5 only. The LLM pricing history page dates these moves across vendors.
How this page was checked
Primary source is the Anthropic pricing page at docs.anthropic.com, fetched on 6 September 2026 and read as extracted text. Every rate above is quoted from its model pricing, batch processing and fast mode tables, and each quote is recorded in the claims file. Cross-checks are the OpenRouter models API (ids anthropic/claude-opus-4 through anthropic/claude-opus-5, anthropic/claude-fable-5 and anthropic/claude-fable-5.1) and the LiteLLM price registry (keys claude-opus-4-20250514, claude-opus-4-1, claude-opus-4-5, claude-opus-4-6, claude-opus-4-7, claude-opus-4-8, claude-opus-5, claude-fable-5, claude-fable-5-1), both fetched the same day and multiplied from per-token to per-million.
Both agreed with the official page on base input, output, cache read and 5 minute cache write for all nine models, including the $0.25 cache read on Fable 5.1. OpenRouter carries no Mythos row; LiteLLM carries claude-mythos-5 and claude-mythos-5-1, and both agree with the official Mythos prices including the $0.25 cache read on 5.1. The worked examples were computed in Python.
Evidence files for this page are published at claude-opus-pricing-by-version-claims.json (every figure with its source quote) and claude-opus-pricing-by-version-calc.txt (the script output for the worked examples).
Questions people search for
How much does the Claude Opus API cost per 1M tokens?
On 6 September 2026 the official Anthropic pricing page lists Claude Opus 5, Opus 4.8, Opus 4.7, Opus 4.6 and Opus 4.5 at $5 per million input tokens and $25 per million output tokens. Opus 4.1 and Opus 4 are listed at $15 input and $75 output and both carry a retired label (4.1 except on Bedrock and Google Cloud, 4 except on Google Cloud).
What is Claude Opus 4.6 pricing?
Claude Opus 4.6 is $5 per million base input tokens, $6.25 for a 5 minute cache write, $10 for a 1 hour cache write, $0.50 per million cache hit tokens and $25 per million output tokens. The batch rate is $2.50 input and $12.50 output. Fast mode is not available on Opus 4.6; such requests run at standard speed and are billed at standard rates.
What is Claude Opus 4.7 API pricing in 2026?
Claude Opus 4.7 costs $5 per million input tokens and $25 per million output tokens on the standard API, and $2.50 and $12.50 on the Batch API. Fast mode is not available on Opus 4.7; the official page says requests with speed set to fast return an error.
How does Opus pricing compare with Sonnet?
Claude Opus 5 is $5 input and $25 output per million tokens, Claude Sonnet 5 is $2 and $10, so Opus 5 costs 2.5x Sonnet 5 on every column. Against Sonnet 4.6 at $3 and $15 the ratio is 1.67x.
Did Anthropic cut Opus prices?
Yes. Opus 4 and Opus 4.1 are listed at $15 input and $75 output. Opus 4.5 and every Opus after it are listed at $5 and $25, one third of the old price. The tier above Opus, Claude Fable 5 and Fable 5.1, is $10 input and $50 output.