Is the include capped?

Yes. One pack is 1B tokens in plus out this month. Unused tokens do not roll over. Need more? Add a pack.

Does paying turn inference on?

No. Request access first. After we enable the workspace, a pack is what you spend.

Who pays? Each engineer?

No. The org Admin buys packs for the workspace. Teammates use that include. They do not buy their own.

Can we share one cue_… key?

No. Each person should create their own key. A shared key makes spend impossible to attribute.

How are the OpenAI and Anthropic numbers calculated?

Published uncached list prices, fetched 2026-08-25. Opus 4.8 is $5 in / $25 out. GPT-5.5 is $5 in / $30 out. We mix 70% input / 30% output, then multiply by the monthly token volume in each scenario. Same volume on Cue is $500 times packs. Those are list math, not invoices. The dated table and chart above have the mix, the other routes, and the source links.

productpricing.md

Pricing

$500 per pack per month = 1B tokens. Add a pack when you need more. Predictable OpEx vs unbounded OpenAI and Anthropic API spend on agent loops.

Pack pricing

$500/ pack / mo1B tokens this period

$0.50/M · $500 = 1B combined in+out. Add a pack when you need more.

pack map

workspace: acme
plan: $0.50/M · $500 = 1B combined in+out this period.
key:  cue_…

What a pack includes

APIs charge per token. We sell packs. $0.50/M · $500 is 1B tokens this month. Need more? Add a pack.

1B tokens in and out, per pack, per month
Unused include does not roll over
Need more? Add a pack.

Uncached 70/30

Run selected frontier-class coding models for 73–78% less. Save more than 90% vs labeled proprietary frontier at the same uncached 70/30 mix.

Cue stays $0.50/M · $500 = 1B combined in+out. List rates fetched 2026-08-25.

Route	List in / out	Uncached 70/30	vs Cue $0.50/M
Cue Cloud ($0.50/M · $500 = 1B combined in+out)	$0.50 / $0.50	$0.50	Cue unit
GLM-5.2 (Z.ai list, uncached)	$1.40 / $4.40	$2.30	Cue 78% lower
DeepSeek V4 Pro 0813 (Peak cache-miss list)	$1.32 / $3.96	$2.11	Cue 76% lower
Kimi K2.7 Code (Kimi list, uncached)	$0.95 / $4	$1.86	Cue 73% lower
Kimi K3 (Kimi K3 is on Cue. Model id ships with the public API list.)	$3 / $15	$6.60	Cue 92% lower
DeepSeek V4 Flash 0731 (Peak cache-miss list)	$0.44 / $1.32	$0.70	No Cue list win
Anthropic · Opus 5 (Different model, uncached)	$5 / $25	$11	Cue 95% lower
Anthropic · Fable 5 (Different model, uncached)	$10 / $50	$22	Cue 98% lower
OpenAI · GPT-5.6 Sol (Different model, uncached)	$4 / $20	$8.80	Cue 94% lower
Anthropic · Opus 4.8 (Different model, uncached)	$5 / $25	$11	Cue 95% lower
OpenAI · GPT-5.5 (Different model, uncached)	$5 / $30	$12.50	Cue 96% lower
OpenAI · GPT-5.6 Terra (Different model, uncached)	$2 / $12	$5	Cue 90% lower
GPT-5.6 Luna (Same $/M. Different model.)	$0.20 / $1.20	$0.50	Same $/M. Different model.

Price vs intelligence

Same model. Same intelligence. 73–78% lower serving price.

Horizontal pairs are the same model, same AA Intelligence Index. Cue at $0.50/M. Hosted points use the uncached 70/30 list blends above. Diamonds are different models. Not a same-model pair. Y is AA Intelligence Index, BenchLM snapshot 2026-08-25. Fetched 2026-08-25.

Cue Cloud
First-party hosted list
Different model
DeepSeek V4 Flash is named on the plot. Peak cache-miss. No Cue list win.

Uncached list prices. 70% input / 30% output. Fetched 2026-08-25. X is first-party hosted list, not a catalog quote. Y is AA Intelligence Index, BenchLM snapshot 2026-08-25. AA Index is a general composite, not a coding-only bench. Do not mix MCP-Atlas onto this axis. Horizontal pairs are the same model. Diamonds are different models. Not a same-model pair. Flash 0731 is a different model from Pro 0813. Flash X is peak cache-miss. Off-peak Flash list can sit under the Cue unit. No Cue list-win stamp on Flash. Luna is on the 70/30 table at the same $/M as Cue. Different model. Not plotted. Cue is $0.50/M · $500 = 1B combined in+out. AA: AA Intelligence Index (BenchLM). Rate sources: OpenAI pricing · Anthropic pricing · Z.ai pricing · DeepSeek pricing · Kimi platform · OpenRouter catalog.

Same tokens. Pack vs meter.

Same pile. Cue is $500 a pack. Labeled OpenAI and Anthropic bill by the million at the 70/30 mix above.

Inside one pack

250M tokens this month. Still one pack.

250Mtokens / mo

Cue Cloud$500

Anthropic$2,750

OpenAI$3,125

$2,250 savings vs Anthropic
$2,625 savings vs OpenAI

One pack · 1B

Full include: 1 pack = 1B tokens (in + out). $500 this period.

1Btokens / mo

Cue Cloud$500

Anthropic$11,000

OpenAI$12,500

$10,500 savings vs Anthropic
$12,000 savings vs OpenAI

Two packs · 2B

Add a pack. $500 × 2. Metered still scales with every turn.

2Btokens / mo

Cue Cloud$1,000

Anthropic$22,000

OpenAI$25,000

$21,000 savings vs Anthropic
$24,000 savings vs OpenAI

Uncached list prices. 70% input / 30% output. Fetched 2026-08-25. Sources: https://developers.openai.com/api/docs/pricing · https://platform.claude.com/docs/en/about-claude/pricing · https://docs.z.ai/guides/overview/pricing · https://api-docs.deepseek.com/quick_start/pricing · https://platform.kimi.ai/ · https://openrouter.ai/models. Cue is $0.50/M · $500 = 1B combined in+out. Caching, batch, peak or off-peak, and harness overhead move real bills. Unused include does not roll over.

Why coding agents punish OpenAI and Anthropic bills

IDE agents re-read repos, call tools, retry, and hold long context. Every turn is more tokens, so OpenAI and Anthropic OpEx grows without a ceiling. On Cue Cloud it is $500 × packs for that include.

Past ~45M blended tokens/mo on Anthropic (sooner on OpenAI at $12.5/MTok), one pack wins on dollars alone, plus open weights on hardware we operate. See models and api.

Pack anatomy

Packs, not people

Price is $500 × packs. Buy the include you need.

One cue_… key

Mint a cue_… key in the console. Use it in CueCode, Cursor, or any OpenAI-compatible client at https://api.cuecloud.io/v1.

Team math you can explain

N packs × $500 = this period’s include. The org Admin picks how many packs.

1 pack · 1B$500/mo

5 packs · 5B$2,500/mo

20 packs · 20B$10,000/mo

Console for humans

Manage packs and keys in the Cue Cloud console at app.cuecloud.io.

FAQ

Is the include capped?: Yes. One pack is 1B tokens in plus out this month. Unused tokens do not roll over. Need more? Add a pack.
Does paying turn inference on?: No. Request access first. After we enable the workspace, a pack is what you spend.
Who pays? Each engineer?: No. The org Admin buys packs for the workspace. Teammates use that include. They do not buy their own.
Can we share one cue_… key?: No. Each person should create their own key. A shared key makes spend impossible to attribute.
How are the OpenAI and Anthropic numbers calculated?: Published uncached list prices, fetched 2026-08-25. Opus 4.8 is $5 in / $25 out. GPT-5.5 is $5 in / $30 out. We mix 70% input / 30% output, then multiply by the monthly token volume in each scenario. Same volume on Cue is $500 times packs. Those are list math, not invoices. The dated table and chart above have the mix, the other routes, and the source links.

Back to explorer