‹ Back to the home page

Prices and benchmarks · checked on 10 November 2026

How much does Claude cost? The same task costs $0.51 or $10.02

How thoroughly the AI works can be set in 5 levels. On the Sonnet 5.5 model the same test task costs $0.51 at the lowest level and $10.02 at the most thorough one — 20 times as much. The result rises from 26 to 55 points. The new Fable 5.2 reaches 54 points even at its lowest level, for $3.10: more than Opus 5.5 at “high” (52 points for $3.46), for 11% less.

In short: AI is billed by the amount of text, in so-called tokens. The price per token says little about the bill, though. What counts is how many tokens a task uses — and that is set mainly by the effort level, not by the price list. All figures come from the official price list, from independent benchmarks and from our own logs, retrieved on 10 November 2026.

20×
difference between the cheapest and the most expensive effort level — in one and the same model (Sonnet 5.5)
$0.41
is what a task costs with Haiku 5.5 at its highest level: 36 points, more than Opus 5.5 at its lowest level for $0.98
+63%
more cost for 0.4 extra points: on Fable 5.2, “xhigh” adds little over “high”
6.3×
as expensive would our own AI work have been if the AI had re-processed known text every time

This page is for anyone who uses AI at work and wants to know what they pay for. What does a token cost? What does a task cost? And which setting really moves the bill? The page does not replace your own costing. It shows the orders of magnitude.

Work out your own cost

Free and without sign-up. Your inputs never leave your browser.

The 4 terms on this page
Token
A text fragment of a few characters. This is what you pay for: per million tokens, separately for text read and text written.
Model
The version of the AI. Claude comes as Fable, Opus, Sonnet and Haiku; the number is the release. Fable is the most expensive, Haiku the cheapest.
Effort level
How thoroughly the model works: 5 levels from “low” to “max”. Higher levels think longer and write more tokens.
Points and tasks
The independent benchmark firm Artificial Analysis gives every model the same test tasks. Points show the result; more is better. Cost per task is the average per test task.

What Teddynews uses

For websites, apps and research: Opus 5.5 on the Ultra plan, Fable 5.2 for the hard cases

Teddynews builds websites, iPhone apps and research reports with Claude Code every day: 97 work sessions and 33 billion tokens from April to September. This is how we decide, based on the figures on this page, as of 10 November 2026.

Subscription instead of pay-per-use

Ultra, formerly “Max 20×”, costs $200 a month. At list prices our use would have cost about $5,800 per 30 days. One weekly limit applies, with no 5-hour limit for verified accounts. Fable may use up to half of it and uses it up faster.

Leave fast mode off

Fast mode is not included in the subscription. It is paid separately from usage credits, at twice the token price, and makes answers faster, not better.

“max” only as an exception

In coding, Opus 5.5 at “max” solves no more tasks than at “xhigh” (59.6% each), Fable 5.2 even fewer (59.6% instead of 61.6%). “Max” costs 49% more on Opus 5.5 and 52% more on Fable 5.2.

Our workModel and effortWhyHow we set it
Everyday: edit a page, write a text, answer a questionOpus 5.5 · medium48 points for $2.45 per task, 7.1 minutes. Solves 52.5% in the coding test, at $7.70 per solved task: cheaper than any other Claude model that solves more than half.nothing, it is the default
Build a website or app, fix bugsOpus 5.5 · highSolves 56.6% of coding tasks instead of 52.5% at “medium”, for 27% more. Checks more edge cases before answering./effort high
Large rework across many filesOpus 5.5 · xhighSolves 59.6%, as much as “max”, for 33% less./effort xhigh
Research with sources, review reports, client textsFable 5.2 · low54 points for $3.10 per task, more than Opus 5.5 at “high” (52 points, $3.46) and faster (6.5 instead of 9.5 minutes). Counts against the Fable half of the weekly limit./model fable, then /effort low
Routine: look things up, rename, summariseHaiku 5.5 · mediumThe fastest Claude model: 222 tokens per second, $0.092 per task./model haiku or as a helper agent
Opus 5.5 gets stuckFable 5.2 · highThe strongest model in the test: 62 points at “high”, 65 at “max”, ahead of every model from every provider in all 6 fields. Costs $9.08 per task and counts against the Fable half of the weekly limit./model fable, then /effort high

Our choice per task. Points and cost per task from the independent benchmark firm Artificial Analysis (Intelligence Index v5.0.1), coding tasks solved from Terminal-Bench 4.0, both retrieved on 10/11/2026. Claude Code starts Fable 5.2 and Opus 5.5 at “medium”, which is why the Fable 5.2 rows name the effort level. On the subscription we pay no token price: fewer tokens there mean more work before a limit applies.

Why Opus 5.5 stays our default: in the overall index it is ahead of Sonnet 5.5 at every level, and on the subscription it does not count against the Fable half. In coding the three are close: at Vals.ai, Fable 5.2 and Opus 5.5 solve 65.2% each, Sonnet 5.5 64.1%.

The biggest lever

The effort level decides: the same AI costs up to 20 times as much

Every recent Claude model has an effort setting that decides how thoroughly it works — 5 levels from “low” to “max”. It affects everything the model writes: the answer, the thinking before it and work steps such as searching or reading files. Through the API, Opus 5.5 and Haiku 5.5 are set to level 2 (“medium”) out of the box, the other models to level 3 (“high”). If you never change it, this default decides your bill.

Cost per task and score, measured by the independent benchmark firm Artificial Analysis (Intelligence Index v5.0.1, 9 evaluations). The score compares models with each other; it is not a grade. Retrieved 10 November 2026.
Model and effortPointsCost per taskCost comparedExtra cost vs. level belowVerdict
Opus 5.5 — recommended by Anthropic
low35$0.98—baseline
medium (default)48$2.45+150%worth it
high52$3.46+41%worth it
xhigh57$6.05+75%worth it
max59$10.04+66%weigh up
Fable 5.2 — new since 09/11/2026, the strongest
low54$3.10—baseline
medium56$3.80+23%worth it
high (default)62$9.08+139%weigh up
xhigh62$14.83+63%skip
max65$26.54+79%weigh up
Sonnet 5.5 — the fast all-rounder
low26$0.51—baseline
medium31$0.78+53%worth it
high (default)42$1.64+110%worth it
xhigh50$4.09+149%worth it
max55$10.02+145%weigh up
Haiku 5.5 — the cheapest
low19$0.044—baseline
medium (default)25$0.092+109%worth it
high30$0.16+74%worth it
xhigh35$0.25+56%worth it
max36$0.41+64%weigh up
The predecessors, still available
Fable 5.1 · high (default)47$7.54—for comparison
Fable 5.1 · max49$12.99—for comparison
Opus 5 · max44$11.82—for comparison
Haiku 4.5 · fixed setting7$0.53—for comparison
On Fable 5.2, “high” is enough: “xhigh” adds 0.4 points for 63% more

Fable 5.2 reaches 61.8 points at “high” for $9.08, and 62.2 points at “xhigh” for $14.83. Only “max” adds 2.3 points again, for a further 79%.

Haiku 5.5 beats Opus 5.5 at its lowest level — for 58% less

Haiku 5.5 at its highest level reaches 36 points for $0.41. Opus 5.5 at its lowest reaches 35 points for $0.98. Haiku writes 12 times as many tokens (226,000 instead of 19,000), but each costs only a fortieth.

The default changes with the model

Through the API, Opus 5.5 and Haiku 5.5 start at “medium”, the other models at “high”. Switch there from Opus 5.5 to Fable 5.2 without setting it, and you work one level higher: 62 instead of 48 points, but 3.7 times the cost per task. Claude Code starts all four current models at “medium”.

How the verdict is set: Up to 20% extra cost per point gained reads “worth it”, above that “weigh up”. If a level gains no point, it reads “skip”. The threshold is our choice, not a measurement — the figures next to it are measured.

Performance

From $9.08 per task, no tested AI scores higher than Fable 5.2

A low price is no use if the result falls short. The independent benchmark firm Artificial Analysis therefore measures score, cost and time on the same tasks. The result: from $9.08 per task, no model with measured cost scores higher than Fable 5.2; at its highest level it reaches 64.6 points, more than any other. Below that the lead changes: Google's Gemini 4 Carbon (60.0 points for $7.76), Opus 5.5 at “xhigh” (57.4 for $6.05), Fable 5.2 at “medium” (56.4 for $3.80). From $0.96 to $3.80, OpenAI leads: GPT-6.1 Terra (51.3 points) and GPT-6.1 Sol (54.6 points for $1.27). Below that, Haiku 5.5 and Meta's Muse Ember. Fable 5.2 leads in all 6 measured fields.

Among Claude models, Haiku 5.5 scores highest up to $0.41, mostly Fable 5.2 from $3.10Score against cost per task. Each line joins the 5 effort levels of one model, from “low” on the left to “max” on the right. How to read it: the higher a point, the better the result; the further left, the cheaper. In between, Sonnet 5.5 at “high” and Opus 5.5 at “medium” lead, and at $6.05 Opus 5.5 at “xhigh” once more. The cost axis grows in multiples: $1 to $10 is as far as $0.10 to $1.
  • Opus 5.5
  • Fable 5.2
  • Sonnet 5.5
  • Haiku 5.5
  • Haiku 4.5 (predecessor, no effort levels)

Opus 5.5: 35 to 59 points for $0.98 to $10.04. Fable 5.2: 54 to 65 points for $3.10 to $26.54. Sonnet 5.5: 26 to 55 points for $0.51 to $10.02. Haiku 5.5: 19 to 36 points for $0.04 to $0.41. Haiku 4.5: 7 points for $0.53. All values are listed in the table in the section “The effort level decides”.

Source: Artificial Analysis, Intelligence Index v5.0.1, retrieved 10/11/2026.

Haiku 5.5 needs 4.0 minutes per task, Opus 5.5 7.1, Fable 5.2 17.5Pure writing time per task, each model at its default level through the API. Waiting time until the first character is not included.
Show values as a table
Pure writing time per task and output speed at the default effort level.
Model and effortTime per taskTokens per second
Haiku 5.5 (medium)4.0 min221.6
Sonnet 5.5 (high)7.0 min139.2
Opus 5.5 (medium)7.1 min103.0
Fable 5.1 (high)13.3 min82.6
Fable 5.2 (high)17.5 min93.0

Source: Artificial Analysis, Intelligence Index v5.0.1, retrieved 10/11/2026.

Fable 5.2 leads in all 6 fieldsPoints per field, each model at its highest level. Each field combines several evaluations. No model from another provider scores higher in any of the 6 fields either.
  • Fable 5.2
  • Opus 5.5
  • Sonnet 5.5
  • Haiku 5.5
Show values as a table
Score per field, each model at its highest effort level.
FieldFable 5.2Opus 5.5Sonnet 5.5Haiku 5.5
Finance and accounting64.759.455.843.6
Strategy and operations67.461.458.139.1
Law63.358.052.238.7
Health and medicine65.457.954.237.5
Engineering72.665.358.542.5
Economics66.560.354.142.7

Source: Artificial Analysis, field scores for Intelligence Index v5.0.1, retrieved 10/11/2026.

For 36 of 48 AI models there is a cheaper choice that is at least as good48 models from 16 providers: every measured effort level for Claude, the benchmark firm’s default selection for the other providers. The higher a point, the better; the further left, the cheaper. The blue line joins the best-value models: no other model is both cheaper and better than they are. If you use a model below the line, you pay too much for the same result. Switch providers on and off; the line is recalculated each time. Not included because the benchmark firm has not measured their cost: Muse Wonder Full (max) (Meta), K2 Horizon 375B A23B (Institute of Foundation Models).
Show or hide providers (16 of 16)
  • Line of best-value models
  • Best-value model
  • Other models

Show values as a table
All models in the comparison, sorted by score. “Open weights” means anyone can download and run the model themselves.
ProviderModel and effortPointsCost per taskTime per taskTokens per second
AnthropicFable 5.2 · max64.6$26.5444.4 min102.8
AnthropicFable 5.2 · xhigh62.2$14.8326.1 min103.4
AnthropicFable 5.2 · high61.8$9.0817.5 min93.0
GoogleGemini 4 Carbon (high)60.0$7.7625.6 min79.9
AnthropicOpus 5.5 · max59.1$10.0425.1 min115.3
AnthropicOpus 5.5 · xhigh57.4$6.0516.2 min106.6
OpenAIGPT-6 Astra (max)56.9$6.2523.1 min37.5
AnthropicFable 5.2 · medium56.4$3.808.1 min88.4
OpenAIGPT-6.1 Sol (max)54.6$1.2720.1 min56.2
AnthropicSonnet 5.5 · max54.5$10.0231.4 min151.3
AnthropicFable 5.2 · low53.9$3.106.5 min92.4
AnthropicOpus 5.5 · high51.8$3.469.5 min105.2
OpenAIGPT-6.1 Terra (max)51.3$0.9613.7 min72.8
AnthropicFable 5.1 · xhigh50.0$10.8519.4 min83.5
AnthropicSonnet 5.5 · xhigh49.9$4.0915.0 min136.0
AnthropicFable 5.1 · max49.2$12.9922.5 min87.4
AnthropicOpus 5.5 · medium48.2$2.457.1 min103.0
AnthropicFable 5.1 · high47.2$7.5413.3 min82.6
KimiKimi K3.1 Pro (open weights)46.5$3.6837.6 min32.0
AnthropicOpus 5 · max44.2$11.8224.6 min80.0
MetaMuse Wonder (max)43.6$6.9512.9 min123.7
AnthropicFable 5.1 · medium43.2$5.569.9 min81.5
AnthropicSonnet 5.5 · high41.5$1.647.0 min139.2
AnthropicFable 5.1 · low38.7$4.477.6 min83.2
AnthropicHaiku 5.5 · max36.5$0.4114.8 min254.2
MetaMuse Ember (max)36.0$0.143.1 min447.7
SpaceXAIGrok 4.7 (xhigh)35.5$7.7725.9 min67.4
AnthropicOpus 5.5 · low34.8$0.983.0 min106.0
AnthropicHaiku 5.5 · xhigh34.6$0.259.6 min235.9
TencentTencent Hy434.6$1.03——
AlibabaQwen3.8 Max (0902)33.7$10.0256.9 min50.0
DeepSeekDeepSeek V4.1 Flash 1017 (max) (open weights)32.6$0.6016.4 min140.9
Z AIGLM-5.3 (max) (open weights)31.7$5.0129.9 min66.3
AnthropicSonnet 5.5 · medium30.8$0.783.6 min141.1
GoogleGemini 3.8 Flash (high)30.8$7.7714.7 min112.3
Thinking MachinesInkling 2 Small30.6$1.1816.7 min149.7
AnthropicHaiku 5.5 · high29.9$0.166.5 min227.3
XiaomiMiMo-V2.6-Pro (open weights)29.9$0.2636.5 min55.2
InclusionAILing 3.1 Flash 1019 (open weights)29.6$3.19——
OpenAIGPT-6 Luna (max)27.9$0.1512.1 min103.9
MistralMistral Large 4 (high) (open weights)26.1$2.0612.7 min108.0
AnthropicSonnet 5.5 · low25.6$0.512.6 min134.6
AnthropicHaiku 5.5 · medium24.6$0.094.0 min221.6
AlibabaQwen3.8 27B (xhigh) (open weights)20.1$3.5340.2 min45.5
AnthropicHaiku 5.5 · low18.6$0.042.1 min217.5
ReflectionBeam 1.0 (open weights)17.7$0.282.1 min277.1
NVIDIANemotron 3.5 Super VL (open weights)13.6$0.525.3 min257.2
AnthropicHaiku 4.57.4$0.534.0 min105.4

Source: Artificial Analysis, Intelligence Index v5.0.1, default selection and model pages, retrieved 10/11/2026. Time per task = pure writing time.

Worked example: coding

Having an app built: a solved task costs $15 to $20 with Opus 5.5 and Sonnet 5.5, $34 with Fable 5.2

The points above are an average across many fields. In coding the differences are especially large. If you let AI build an app, you also pay for attempts that fail. What counts is the cost per solved task. In the independent coding benchmark Terminal-Bench 4.0, Sonnet 5.5 at its highest level solves the most tasks among Claude models: 63.6%, for $19.87 per solved task. Fable 5.2 at “xhigh” solves 61.6% for $33.78, Opus 5.5 at “xhigh” 59.6% for $14.73. Fable 5.1 needs $28.21, Opus 5 $39.37. The highest score of all measured models goes to the open-weights Kimi K3.1 Pro with 67.2%.

How it was measured: Terminal-Bench 4.0 consists of 66 tasks from software development, machine learning, science, IT operations, security, hardware and media. Example: move a running application from MySQL to PostgreSQL without downtime; every response must stay the same, and response time may rise by at most 10 milliseconds. The AI works on its own through a command line, for up to 8 hours per task. A task counts as solved only if all automated tests pass. Artificial Analysis runs each task 3 times and measures cost, tokens and time.

Sonnet 5.5 solves the most, Opus 5.5 is cheapest per solutionEach model at the effort level with its best result. For Opus 5.5, “xhigh” and “max” reach the same 59.6%; “xhigh” is cheaper. Fable 5.2 and Fable 5.1 solve more at “xhigh” than at “max”. Opus 5 was only measured at “max”. One bar per model for each measure: more is better for tasks solved, less is better for cost, tokens written and time.
  • Sonnet 5.5 (max)
  • Fable 5.2 (xhigh)
  • Opus 5.5 (xhigh)
  • Fable 5.1 (xhigh) and Opus 5 (max), predecessors
Show values as a table
Terminal-Bench 4.0, values per task. Cost per solved task = cost per task divided by the share of tasks solved.
Model and effortSolvedCost per taskCost per solved taskTokens writtenTime
Sonnet 5.5 (max)63.6%$12.65$19.87382,42942.1 min
Fable 5.2 (xhigh)61.6%$20.81$33.78259,77141.9 min
Opus 5.5 (xhigh)59.6%$8.78$14.73169,49726.5 min
Fable 5.1 (xhigh)55.1%$15.53$28.21158,35231.6 min
Opus 5 (max)49.0%$19.29$39.37205,11042.7 min

Source: Artificial Analysis, Terminal-Bench 4.0 in Intelligence Index v5.0.1, values per task, retrieved 10/11/2026. Cost per solved task calculated by us.

Even “medium” beats Opus 5 at its highest level — for 21% of the costOne line per model through its 5 effort levels: the higher, the more tasks solved; the further left, the cheaper. Anthropic states that Opus 5.5 at its default beats Opus 5 at maximum effort for about a fifth of the cost. The independent measurement confirms it. Fable 5.2 solves more than Opus 5.5 at each level from “low” to “xhigh”, but costs more each time.
  • Opus 5.5
  • Fable 5.2
  • Sonnet 5.5
  • Opus 5 (only measured at max)

Opus 5.5 “medium”: 52.5% solved for $4.04 per task. Fable 5.2 “low”: 54.5% for $4.34, “xhigh”: 61.6% for $20.81. Sonnet 5.5 “max”: 63.6% for $12.65. Opus 5 “max”: 49.0% for $19.29. Haiku 5.5 solves at most 32.8% for $0.65 and is therefore outside the chart.

Show values as a table
Terminal-Bench 4.0 per effort level: share of tasks solved and cost per task.
Model and effortSolvedCost per task
Opus 5.5 (low)31.3%$2.08
Opus 5.5 (medium)52.5%$4.04
Opus 5.5 (high)56.6%$5.12
Opus 5.5 (xhigh)59.6%$8.78
Opus 5.5 (max)59.6%$13.11
Fable 5.2 (low)54.5%$4.34
Fable 5.2 (medium)55.1%$5.25
Fable 5.2 (high)59.6%$13.47
Fable 5.2 (xhigh)61.6%$20.81
Fable 5.2 (max)59.6%$31.59
Sonnet 5.5 (low)20.7%$1.28
Sonnet 5.5 (medium)29.8%$1.58
Sonnet 5.5 (high)43.9%$2.19
Sonnet 5.5 (xhigh)57.1%$6.08
Sonnet 5.5 (max)63.6%$12.65
Opus 5 (max)49.0%$19.29

Source: Artificial Analysis, Terminal-Bench 4.0, retrieved 10/11/2026. Vendor claim: Anthropic, Opus 5.5 announcement of 22/09/2026.

“xhigh” is enough: 33% cheaper, same number solved

Opus 5.5 solves 59.6% at both “xhigh” and “max”. “Max” costs $13.11 instead of $8.78 per task and writes 253,000 instead of 169,000 tokens. On Fable 5.2, “max” even solves fewer than “xhigh”.

Sonnet 5.5 solves the most — and writes the most

At its highest level Sonnet 5.5 solves 63.6% of the tasks and writes 382,000 tokens per task for it, more than twice as many as Opus 5.5 at “xhigh” (169,000). Because each token costs half as much, it ends up at $12.65 instead of $8.78 per task.

Sensitive requests are included

Requests the provider flags as sensitive, such as IT security tasks, are answered by an older model on Opus 5.5, Sonnet 5.5 and the Fable models. The benchmark firm measures them exactly this way and labels it (“with fallback”); the values include these answers.

4 measurements: Fable 5.2, Opus 5.5 and Sonnet 5.5 are close together

Share of tasks solved in Terminal-Bench 4.0. The values differ because each measurement gives the AI a different working program. Which of the three leads depends on the measurement. Retrieved 10/11/2026.
SourceAI working programFable 5.2Opus 5.5Sonnet 5.5Fable 5.1
Artificial Analysis (independent)mini-swe-agent61.6%59.6%63.6%55.1%
Vals.ai (independent)Mini-SWE-agent65.2%65.2%64.1%58.1%
Snorkel (independent)Claude Code—64.8%61.8%57.9%
Anthropic (vendor)not stated—66.4%70.6%55.8%

“—” means: not published there. Artificial Analysis: Fable 5.2, Opus 5.5 and Fable 5.1 at “xhigh”, Sonnet 5.5 at “max”, each at its best level. Snorkel measures all at “max”. Anthropic gives no effort level for Sonnet 5.5; the Vals.ai values apply to the tester's standard setting. Vals.ai measured Fable 5.2 on 09/11/2026; Snorkel and Anthropic do not yet give a value for Fable 5.2 in this test.

Base prices

What the AI writes costs 5 times as much as what it reads

All the costs above come from a handful of prices. Billing is per token — a piece of text, around 4 characters in English. Text the AI reads (your question, attached documents, the conversation so far) is much cheaper than text it writes. All prices per million tokens in US dollars, retrieved on 10 November 2026.

Prices per million tokens. Text that is needed several times, such as the conversation so far, is stored once (the “cache”) and then read again for a fraction of the price. Storing costs a little more than newly entered text, once. “Max. text” is how many tokens the model can handle at once. Haiku 5.5 costs five times as much for requests over 100,000 tokens (100K).
ModelNewly enteredWritten by the AIStored for laterReusedMax. text
Fable 5.2$8.00$40.00$10.00$0.201M
Opus 5.5$4.00$20.00$5.00$0.201M
Sonnet 5.5$2.00$10.00$2.50$0.101M
Haiku 5.5, up to 100K$0.10$0.50$0.125$0.011M
Haiku 5.5, over 100K$0.50$2.50$0.625$0.051M
Fable 5.1$10.00$50.00$12.50$0.251M
Opus 5$5.00$25.00$6.25$0.501M
Sonnet 5$2.00$10.00$2.50$0.201M
Haiku 4.5$1.00$5.00$1.25$0.10200K
Fable 5.2, fast mode$16.00$80.00$20.00$0.401M
Opus 5.5, fast mode$8.00$40.00$10.00$0.401M

2 discounts come on top: if you collect tasks and have them processed as a batch, you pay half. If you need the same text several times, reading it again costs only 10% of the input price. On Opus 5.5 and Sonnet 5.5 it is 5%, on Fable 5.2 and Fable 5.1 just 2.5%. Fast mode, on Fable 5.2 and Opus, delivers answers up to 2.5 times faster at twice the price; it does not make the model smarter.

Cost traps

4 cost traps no price list mentions

Besides price and effort level, 4 things push up the bill without anyone noticing.

Same page, more tokens

Models from Claude 4.7 onwards split the same text into about 30% more tokens than their predecessors. An identical price per token therefore does not mean the same price per page of text. Anyone comparing old and new models has to allow for this.

Switching models changes the effort

Without a setting, Opus 5.5 works one level lower than Opus 5. Sonnet 5.5 has recalibrated levels: the same level thinks a different amount than on Sonnet 5. Set the effort level explicitly whenever you change models.

Long requests cost five times as much on Haiku 5.5

Haiku 5.5 costs $0.10 per million input tokens as long as a request has at most 100,000 tokens. Above that, the whole request costs five times as much, including text read from the store. Long documents and long conversations make the cheapest model much more expensive.

Reuse saves the most

In long sessions the AI reads known text from a store instead of processing it again. Without that, our own work would have cost 6.3 times as much. Important: changing the effort for the whole request or switching to fast mode empties this store, and you pay full price again. Changing the effort for single messages only keeps it — available on Fable 5.2, Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5 and Haiku 5.5.

Our own figures

Our own AI use: 6.3 times cheaper by reusing known text

How big these levers are in practice is shown by our own use. Teddynews uses Claude every day. We analysed all our work sessions from 20 April to 25 September 2026. At the price list of 26 September 2026 this use would have cost $30,584 — that is what you pay if every request is billed by usage. We ourselves pay a subscription with a fixed monthly price.

The main saving comes from reuse. In long conversations the program resends the whole history with every new question. The AI then reads text it already knows from a store instead of processing it again — for 2.5 to 10% of the normal price. Without this reuse the same work would have cost $192,754, 6.3 times as much.

97
work sessions on 133 of 159 days, with 72,192 AI responses
33.2 bn
tokens processed (text fragments of a few characters), 114 million of them written by the AI itself
$30,584
is what the use would have cost at the price list of 26/09/2026 — we pay a fixed-price subscription
6.3×
as expensive would it have been if the AI had re-processed known text every time
Reused text: 97% of the volume, only 54% of the costHow the $30,584 split across the 4 items every Claude bill lists. What the AI writes itself is only 0.34% of the volume, yet 10% of the cost.
  • Reused (read from store)
  • Stored for later
  • Written by the AI
  • Newly entered
Show values as a table
The 4 items on the bill: share of the volume (tokens) and of the cost. Amounts rounded; the total is formed from unrounded values.
ItemTokensShare of tokensCostShare of cost
Reused (read from store)32,159.3M96.77%$16,50353.96%
Stored for later956.5M2.88%$10,86635.53%
Written by the AI113.8M0.34%$3,19710.45%
Newly entered3.6M0.01%$190.06%
Total33,233.2M100%$30,584100%

Source: our own Claude Code logs, 20/04–25/09/2026, valued with the provider's price list of 26/09/2026.

Our use per month: 9 times as much in September as in MayIn dollars at the price list of 26/09/2026 — what the use would have cost without a subscription. April counts from the 20th, September up to the 25th.

Total 20 April to 25 September: $30,584 · 72,192 AI responses · 33.2 bn tokens

Show values as a table
Our use per month at the price list of 26/09/2026 (Vienna time).
MonthAI responsesTokensAt list price
April3,5641.24 bn$1,093
May8,8981.22 bn$1,023
June13,6376.90 bn$6,577
July11,5515.94 bn$6,748
August12,6466.57 bn$5,924
September21,89611.37 bn$9,219
Total72,19233.23 bn$30,584

Source: our own Claude Code logs, 20/04–25/09/2026, valued with the provider's price list of 26/09/2026.

How we counted: The basis is the Claude Code logs on the Teddynews work computer; each separate session log counts as one session. Every AI response counts once, even if the log writes it several times; the final value applies. Each response is valued at the price list of its model of 26/09/2026; storing text at the rate for the storage time shown in the log. Not included: sessions before 20 April 2026 (no logs exist for them), other tools and other computers. For 0.36% of the amount the log does not say whether the fast mode at twice the price was used; we counted the normal price.

Calculator

Calculator: what your AI use costs on 8 models

Estimate your own use with the same 4 items as in our own bill. The example is a longer working session. Everything is calculated in your browser — no input is transmitted.

Everything read for the first time
Known text read from the store — costs 2.5 to 10% of the input price, depending on the model
One-off storage, 1.25 times the input price
Answer, thinking and work steps together
Cost of this session —
The same session with other models

Questions

9 answers: what Claude costs per month and per task

How much does Claude cost per month?

Free costs nothing. Pro costs $20 a month, or $17 with annual billing ($200 up front). Max costs $100 for 5 times the Pro usage. Ultra costs $200: 10 times the weekly Pro usage and, for verified accounts, no 5-hour limit. Prices exclude VAT; in Germany 19% is added, in Austria 20%. As shown on the provider’s pricing page, retrieved 12/11/2026.

Does this also apply to subscriptions such as ChatGPT Plus or Claude Pro?

No. A subscription has a fixed price; you do not pay per task. The figures on this page apply to usage-based billing, as it arises when AI is built into your own software. For subscribers they are still a yardstick: they show how quickly a usage allowance is used up.

Why is the most expensive level not always the best?

Because extra thinking stops adding value at some point. For Fable 5.2 this point lies between “high” and “xhigh”: 0.4 more points for 63% more cost. For Opus 5.5 the top level still adds 2 points but costs 66% more. For very long tasks it can still make sense — then for stamina, not for quality.

What changes with Fable 5.2 compared with Fable 5.1?

3 things. The token price is 20% lower: $8 instead of $10 per million input tokens, $40 instead of $50 for output. In the overall index Fable 5.2 scores higher at every level, 54 instead of 39 at “low”; at “low” and “medium” it also costs less per task, from “high” upwards more, because it writes more. And there is a fast mode at twice the price. Fable 5.1 remains available for now.

Which Claude model is the cheapest?

Haiku 5.5: $0.10 per million input tokens and $0.50 for output, as long as a request stays under 100,000 tokens. Per test task it costs $0.04 to $0.41 and reaches 36 points at its highest level, more than Opus 5.5 at its lowest. It is not enough for demanding coding: in the coding test it solves at most 33% of the tasks, Sonnet 5.5 up to 64%.

How much does a token cost on Claude?

A token is a piece of text, about 4 characters in English according to the provider. Billing is per million: on Opus 5.5, 1 million tokens read cost $4 and 1 million written cost $20. On Haiku 5.5 it is $0.10 and $0.50, as long as a request stays under 100,000 tokens. Models from Claude 4.7 onwards split the same text into about 30% more tokens.

How do I get my own usage figures?

Every bill lists the 4 items from the calculator separately: newly entered, reused, stored for later, written by the AI. Developer tools usually have their own command for it, web interfaces a usage page in the account.

Are the scores from Anthropic itself?

No. They come from Artificial Analysis, an independent benchmark firm that rates models using a fixed method of 9 evaluations. The prices come from the official price list. Both are linked below with their retrieval date.

Why does Teddynews publish this?

Because we use AI in our own tools and had to do the same calculations. Who is behind it and what they sell is stated openly in the section Who writes this page.

Disclosure

Who writes this page — and what they sell

Teddynews builds AI tools — including AI agents for businesses (German) — and calculates with the same prices. Our own figures in the section “Our own AI use” come from this work. The page costs nothing, collects no address and recommends no provider.

Anyone who publishes something like this usually has something to sell. That is true here too. So it is stated openly, with its real status:

Teddy-LMC

Your documents become a tool that pre-sorts enquiries — instead of a PDF to download.

In preparation. €149 is a proposal, not yet billable.

See Teddy-LMC

Teddy carousel

10 Instagram cards a month, built from your reader's own words. With a forecast of which ones get shared.

€149 once, no subscription. See three sample cards for free.

See the Teddy carousel

More overviews with sources

Made the same way as this page: electricity and AI, the training market, the reasons for AI scepticism. All with sources.

Free, without sign-up.

AI as a tool · CO₂ and AI

No newsletter, no call-back, no deadline. If you have a question, write to office@teddynews.at. A person answers, not a queue.

Where the figures come from

Sources: 10 references with origin and retrieval date

All prices are official — they come from the provider. Scores, cost, tokens and time per task are independently measured — measured by independent benchmark firms, not confirmed by the provider; vendor claims are named as such. The figures in the section “Our own AI use” are our own count — counted from our logs. No value on this page is estimated; calculated values such as the cost per solved task are marked as such.

How this page was made: Research and text were created with AI assistance and checked by Teddynews (disclosure under Art. 50 of the EU AI Act). Data as of 10/11/2026; prices and measurements were retrieved the same day. Our own usage figures date from 26/09/2026.

For the cost per task and effort level there is currently no second independent measurement; Artificial Analysis is the only benchmark firm measuring this way. Prices change. This page therefore gives the retrieval date for every figure. If you use it as the basis for a costing, please check the current state at the linked source.

‹ Back to the home page
Buy Teddy a Coffee