Savings

What you save when Dex makes the decisions.

Let Dex answer the typed questions about every ticket, message or tool call, and pay an LLM only for the text it writes.

1,000 support tickets, 14 decisions eachrecorded 2026-09-25

  • Dex, at the Fierce plan rateEUR 0.110.153 s per ticket
  • GPT-4.1 mini, the cheapest LLM we racedEUR 0.661.52 s per ticketDex is 5.9× cheaper
  • Claude Fable 5.1EUR 51.1111.56 s per ticketDex is 463× cheaper3.2 h less waiting per 1,000 tickets

Tokens times list price, with LLM input and output priced separately; Dex output is free. Dex at the Fierce plan rate (EUR 0.09 per million input tokens). At pay as you go (EUR 1.00 per million input tokens) the same tickets cost EUR 1.21 with Dex, more than GPT-4.1 mini. LLM list prices read 2026-09-25; USD at 0.858664 EUR. Times are medians of 10 recorded runs, Dex over the network like the LLM calls. How we measured

At a glance

Six scenarios at 50,000 events a month.

Monthly cost with Claude Fable 5.1 alone, with Dex making the decisions and Claude Fable 5.1 writing, and with Dex plus Claude Sonnet 5. With the same LLM, the saving is the decision calls Dex takes over, so scenarios with the same decision block save the same amount. Default assumptions; change them in the scenarios below.

Monthly cost of six scenarios at 50,000 events a month
ScenarioClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Customer chatbotEUR 3,239.10EUR 707.71saves EUR 2,531.38 (78%)EUR 160.74saves EUR 3,078.35 (95%)
Support desk triageEUR 3,070.58EUR 539.20saves EUR 2,531.38 (82%)EUR 127.04saves EUR 2,943.54 (95%)
AI agent with tool callsEUR 1,486.35EUR 600.42saves EUR 885.93 (59%)EUR 136.74saves EUR 1,349.60 (90%)
Content moderationEUR 1,182.38EUR 165.89saves EUR 1,016.49 (85%)EUR 52.31saves EUR 1,130.08 (95%)
Lead qualificationEUR 1,535.29EUR 333.12saves EUR 1,202.17 (78%)EUR 85.82saves EUR 1,449.47 (94%)
RAG question routingEUR 2,452.34EUR 1,566.42saves EUR 885.93 (36%)EUR 329.94saves EUR 2,122.40 (86%)

Next to GPT-4.1 mini, the cheapest LLM we raced, the picture is different: Dex + GPT-4.1 mini saves money in 3 of 6 scenarios at 50,000 events a month and in 6 of 6 at 500,000. Pick GPT-4.1 mini in the scenarios to see every number.

Integration scenarios

Six ways to put Dex in front of an LLM.

Each scenario says what Dex decides and what the LLM writes. Decision tokens are measured in our recorded races; the text an LLM writes is an estimate you can change. Monthly totals for 5,000, 50,000 and 500,000 events.

measured in our recorded races estimate, change it

01 Customer chatbot

Dex decides
intent, sentiment, escalation and guardrail checks on every turn
The LLM does
writes the reply on the turns that need one

Decision block: our support ticket triage race, 14 questions. 1,212 billable tokens per call for Dex; 2,892 input and 612 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
Customer chatbot: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 6.06EUR 323.91EUR 74.43saves EUR 249.48 (77%)EUR 19.73saves EUR 304.18 (93%)
50,000Dex: Hot, EUR 24.00EUR 3,239.10EUR 707.71saves EUR 2,531.38 (78%)EUR 160.74saves EUR 3,078.35 (95%)
500,000Dex: Fierce + 1 reset a week, EUR 81.25EUR 32,390.95EUR 6,918.36saves EUR 25,472.59 (78%)EUR 1,448.67saves EUR 30,942.28 (95%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 60.6 million tokens on HotEUR 0.00EUR 24.00EUR 24.00
LLM decisions, 50,000 callsEUR 2,555.38EUR 0.00EUR 0.00
LLM replies, 32,500 callsEUR 683.71EUR 683.71EUR 136.74
Total a monthEUR 3,239.10EUR 707.71EUR 160.74

02 Support desk triage

Dex decides
team, category, urgency and sentiment for every ticket
The LLM does
drafts a reply for the tickets routed to one

Decision block: our support ticket triage race, 14 questions. 1,212 billable tokens per call for Dex; 2,892 input and 612 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
Support desk triage: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 6.06EUR 307.06EUR 57.58saves EUR 249.48 (81%)EUR 16.36saves EUR 290.69 (94%)
50,000Dex: Hot, EUR 24.00EUR 3,070.58EUR 539.20saves EUR 2,531.38 (82%)EUR 127.04saves EUR 2,943.54 (95%)
500,000Dex: Fierce + 1 reset a week, EUR 81.25EUR 30,705.82EUR 5,233.23saves EUR 25,472.59 (82%)EUR 1,111.65saves EUR 29,594.18 (96%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 60.6 million tokens on HotEUR 0.00EUR 24.00EUR 24.00
LLM decisions, 50,000 callsEUR 2,555.38EUR 0.00EUR 0.00
LLM drafts, 25,000 callsEUR 515.20EUR 515.20EUR 103.04
Total a monthEUR 3,070.58EUR 539.20EUR 127.04

03 AI agent with tool calls

Dex decides
allow, ask or block for every tool call, instead of a second LLM call
The LLM does
runs the agent's own step that proposes the tool call

Decision block: our agent guardrail race, 7 questions. 662 billable tokens per call for Dex; 1,752 input and 72 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
AI agent with tool calls: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 3.31EUR 148.63EUR 61.27saves EUR 87.36 (58%)EUR 14.90saves EUR 133.73 (89%)
50,000Dex: Base + 1 reset a week, EUR 20.82EUR 1,486.35EUR 600.42saves EUR 885.93 (59%)EUR 136.74saves EUR 1,349.60 (90%)
500,000Dex: Fierce, EUR 39.00EUR 14,863.47EUR 5,834.98saves EUR 9,028.49 (60%)EUR 1,198.20saves EUR 13,665.28 (91%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 33.1 million tokens on Base + 1 reset a weekEUR 0.00EUR 20.82EUR 20.82
LLM decisions, 50,000 callsEUR 906.75EUR 0.00EUR 0.00
LLM agent steps, 50,000 callsEUR 579.60EUR 579.60EUR 115.92
Total a monthEUR 1,486.35EUR 600.42EUR 136.74

04 Content moderation

Dex decides
every item, with min_confidence so it abstains when unsure
The LLM does
takes a second look at the items Dex abstained on

Decision block: our marketplace moderation race, 9 questions. 910 billable tokens per call for Dex; 2,219 input and 107 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
Content moderation: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 4.55EUR 118.24EUR 18.74saves EUR 99.50 (84%)EUR 7.38saves EUR 110.86 (93%)
50,000Dex: Hot, EUR 24.00EUR 1,182.38EUR 165.89saves EUR 1,016.49 (85%)EUR 52.31saves EUR 1,130.08 (95%)
500,000Dex: Fierce + pay as you go above the allowance, EUR 65.00EUR 11,823.80EUR 1,483.86saves EUR 10,339.95 (87%)EUR 348.05saves EUR 11,475.75 (97%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 45.5 million tokens on HotEUR 0.00EUR 24.00EUR 24.00
LLM decisions, 50,000 callsEUR 1,182.38EUR 0.00EUR 0.00
LLM reviews, 6,000 callsEUR 0.00EUR 141.89EUR 28.31
Total a monthEUR 1,182.38EUR 165.89EUR 52.31

05 Lead qualification

Dex decides
fit, intent and priority for every inbound lead
The LLM does
writes a follow-up email for the qualified leads

Decision block: our lead qualification race, 11 questions. 917 billable tokens per call for Dex; 2,276 input and 116 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
Lead qualification: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 4.59EUR 153.53EUR 35.50saves EUR 118.03 (76%)EUR 10.77saves EUR 142.76 (92%)
50,000Dex: Hot, EUR 24.00EUR 1,535.29EUR 333.12saves EUR 1,202.17 (78%)EUR 85.82saves EUR 1,449.47 (94%)
500,000Dex: Fierce + pay as you go above the allowance, EUR 68.50EUR 15,352.91EUR 3,159.69saves EUR 12,193.22 (79%)EUR 686.74saves EUR 14,666.17 (95%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 45.8 million tokens on HotEUR 0.00EUR 24.00EUR 24.00
LLM decisions, 50,000 callsEUR 1,226.17EUR 0.00EUR 0.00
LLM follow-ups, 15,000 callsEUR 309.12EUR 309.12EUR 61.82
Total a monthEUR 1,535.29EUR 333.12EUR 85.82

06 RAG question routing

Dex decides
which collection to search, and whether the question can be answered, before retrieval
The LLM does
writes the answer from the retrieved passages

Decision block: our agent guardrail race, the smallest block we measured, 7 questions. 662 billable tokens per call for Dex; 1,752 input and 72 output tokens for Claude Fable 5.1 measured in our recorded races

Assumptions
RAG question routing: monthly cost at three volumes
Events a monthClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
5,000Dex: Pay as you go, EUR 3.31EUR 245.23EUR 157.87saves EUR 87.36 (35%)EUR 34.22saves EUR 211.01 (86%)
50,000Dex: Base + 1 reset a week, EUR 20.82EUR 2,452.34EUR 1,566.42saves EUR 885.93 (36%)EUR 329.94saves EUR 2,122.40 (86%)
500,000Dex: Fierce, EUR 39.00EUR 24,523.44EUR 15,494.95saves EUR 9,028.49 (36%)EUR 3,130.19saves EUR 21,393.25 (87%)
How 50,000 events a month add up
How 50,000 events a month add up
PartClaude Fable 5.1 onlyDex + Claude Fable 5.1Dex + Claude Sonnet 5
Dex decisions, 33.1 million tokens on Base + 1 reset a weekEUR 0.00EUR 20.82EUR 20.82
LLM decisions, 50,000 callsEUR 906.75EUR 0.00EUR 0.00
LLM answers, 40,000 callsEUR 1,545.60EUR 1,545.60EUR 309.12
Total a monthEUR 2,452.34EUR 1,566.42EUR 329.94

Per call

Compare one kind of call, LLM by LLM.

Pick a recorded workflow or enter your own call, set the volume and the plan. Every LLM is priced from its measured input and output tokens at its list price.

Monthly cost of one kind of call

Dex plan
DexEUR 25.20 a month on Hot + pay as you go above the allowance

1,212 billable tokens per call, 121.2 million a month; 0.153 s per call; agreement 79%

This is the plan that fits this volume.

Monthly cost of one kind of call
LLMList price per million tokensLLM per monthDex savesWaiting saved a monthAgreement
GPT-4.1 mini1,569 in, 87 outUSD 0.40 input, 1.60 outputsource, read 2026-09-25EUR 65.84EUR 40.64 (61%)38.1 h86%
GPT-5.4 mini1,568 in, 93 outUSD 0.75 input, 4.50 outputsource, read 2026-09-25EUR 136.91EUR 111.71 (81%)23.9 h78%
Claude Haiku 4.52,300 in, 97 outUSD 1 input, 5 outputsource, read 2026-09-25EUR 239.14EUR 213.94 (89%)46.6 h79%
GPT-4.11,569 in, 85 outUSD 2 input, 8 outputsource, read 2026-09-25EUR 327.84EUR 302.64 (92%)48.0 h93%
GPT-5.3 chat1,568 in, 91 outUSD 1.75 input, 14 outputsource, read 2026-09-25EUR 345.01EUR 319.81 (92%)33.1 h86%
Claude Sonnet 52,890 in, 138 outUSD 2 input, 10 outputsource, read 2026-09-25EUR 614.80EUR 589.60 (95%)58.8 h99%
Claude Opus 5.52,892 in, 143 outUSD 4 input, 20 outputsource, read 2026-09-25EUR 1,238.88EUR 1,213.68 (97%)83.8 h93%
Claude Fable 5.12,892 in, 612 outUSD 10 input, 50 outputsource, read 2026-09-25EUR 5,110.77EUR 5,085.57 (99%)317 h93%

Agreement is the share of answers equal to the reference answers recorded with the race, for this one workflow. Dex: 79%.

What these numbers do not include

  • Retries, timeouts and failed LLM calls. Dex refunds a call that fails on our side.
  • Discounts on the LLM side: batch APIs, prompt caching and negotiated rates. We use standard list prices, global routing, no caching.
  • Writing. Dex answers typed questions only (pick, rate, check), so every scenario keeps an LLM wherever text is written.
  • One call instead of two. An LLM-only setup can ask for the decisions and the reply in one call and save some input tokens; we price the decision call as we measured it.
  • Quality. It differs per use case: agreement with the reference answers per workflow is in the race table, and the measured quality of the Dex model is in the docs. Dex's rate questions are in beta.
  • Estimates marked [E]: the tokens of the text an LLM writes are our assumptions, not measurements.
  • VAT, retrieval and embedding costs, hosting of your own code and engineering time.

Sources

Every price, every measurement.

LLM list prices as the race files record them, with the page each was read from and the date. Dex prices from our pricing page.

LLM list prices per million tokens
LLMInputOutputSource
Claude Fable 5.1USD 10USD 50source, read 2026-09-25
Claude Opus 5.5USD 4USD 20source, read 2026-09-25
Claude Sonnet 5USD 2USD 10source, read 2026-09-25
Claude Haiku 4.5USD 1USD 5source, read 2026-09-25
GPT-5.3 chatUSD 1.75USD 14source, read 2026-09-25
GPT-4.1USD 2USD 8source, read 2026-09-25
GPT-5.4 miniUSD 0.75USD 4.50source, read 2026-09-25
GPT-4.1 miniUSD 0.40USD 1.60source, read 2026-09-25

USD converted at 0.858664 EUR per USD (Azure Retail Prices API rate, as used in docs/pricing-model.md, read 2026-09-25). Claude models ran with thinking at its lowest setting where it cannot be switched off; thinking tokens are billed as output and are in the counts.

Dex prices
PlanPer monthEffective rate
Pay as you gononeEUR 1.00 per million input tokens
BaseEUR 9.99 a monthEUR 0.50 per million input tokens
HotEUR 24.00 a monthEUR 0.20 per million input tokens
FierceEUR 39.00 a monthEUR 0.09 per million input tokens

Pricing

Recorded workflows
WorkflowQuestionsDex tokens per callRecorded
EU SaaS support ticket triage141,2122026-09-25
Inbound B2B lead qualification119172026-09-25
Dutch marketplace comment moderation99102026-09-25
AI agent guardrail check before a tool call76622026-09-25

Questions

Savings questions.

How much does Dex save?

It depends on the LLM you use today and on how much of your traffic is decisions. In our six scenarios at 50,000 events a month, Dex plus Claude Fable 5.1 costs 36% to 85% less than Claude Fable 5.1 alone, and Dex plus Claude Sonnet 5 costs 86% to 95% less. For 1,000 support tickets, Dex costs EUR 0.11 at the Fierce plan rate, against EUR 51.11 with Claude Fable 5.1.

Is Dex always cheaper than an LLM?

No. Next to GPT-4.1 mini, the cheapest LLM we raced, Dex saves money in 3 of 6 scenarios at 50,000 events a month and in 6 of 6 at 500,000. At pay as you go, 1,000 support tickets cost EUR 1.21 with Dex and EUR 0.66 with GPT-4.1 mini. The saving grows with volume, because a plan's weekly allowance makes each token cheaper.

Where do the numbers come from?

Decision tokens and times come from our recorded races of 2026-09-25: the same questions answered by Dex and by 8 LLMs, 10 runs each. LLM costs are those tokens times the providers' list prices, input and output priced separately. The tokens of text an LLM writes are estimates, marked [E], and you can change them.

Why do the Dex columns still include an LLM?

Dex answers typed questions: pick, rate and check. It does not write text. In every scenario an LLM still writes the replies, drafts, emails, agent steps or answers; Dex takes over the decisions around them, so the LLM runs less often or can be a smaller model.

Which Dex plan do the scenarios use?

The plan that fits each month: pay as you go, or the plan whose weekly allowance, resets and use beyond the allowance cost least for that volume, with the same rules as the pricing page. The headline numbers use the Fierce plan rate and show pay as you go next to it.