Skip to content

One bill. Every model.

One bill for every lab. Rates are the lab's list. Spend caps, roles, and keys are in the product, not a plan.

billing

Billing

[BALANCE]

747.29

tokens remaining

token balance152.71 / 900
topped up
1,000
spent
165.55
per-request ceiling
12
$ / Token
1.00
tokens / min
2,000,000
auto-reload
20%
BYOK
2 providers · not debited
[LIST PRICE][NO MARKUP]

List prices.

MODELCONTEXTIN / 1MOUT / 1M
Qwenqwen/qwen3.8-27b262.1k0.060.24
Googlegoogle/gemini-2.5-flash-lite1.0M0.100.40
Mistralmistral/mistral-small-4262.1k0.150.60
Mistralmistral/ministral-3-8b262.1k0.150.15
OpenAIopenai/gpt-5.6-luna1.0M0.201.20
OpenAIopenai/gpt-5.4-nano1.0M0.201.25
Googlegoogle/gemini-3.5-flash-lite1.0M0.302.50
Googlegoogle/gemini-2.5-flash1.0M0.302.50
Metameta/muse-glimmer-30b131.1k0.301.20
DeepSeekdeepseek/deepseek-v4-flash1.0M0.441.32
Mistralmistral/mistral-large-3262.1k0.501.50
OpenAIopenai/gpt-5.4-mini1.0M0.754.50
Anthropicanthropic/claude-haiku-4-5200.0k1.005.00
xAIxai/grok-build-0.1256.0k1.002.00
Z.aizai/glm-5-turbo202.8k1.204.00
xAIxai/grok-4.3256.0k1.252.50
Metameta/muse-spark-1.21.0M1.254.25
DeepSeekdeepseek/deepseek-v4-pro1.0M1.323.96
Z.aizai/glm-5.21.0M1.404.40
1–19 of 37

Per million tokens, at the primary upstream. Extra upstreams have their own prices. A dash means no cached rate. Reasoning bills as completion. Cached reads are the lab's cache, at their rate.

Project caps.

Each project has a spend cap. Each call has a ceiling. Go over either and the request is refused before it runs.

projects

Billing

The wallet is the maximum. Cap a project below it.
[ALLOC]

Allocation

wallet
1,000
allocated
1,000
unallocated
0
  • latticeceiling 12 · 600 rpm
    cap152.71 / 900
  • hearthceiling 12 · 600 rpm
    cap12.84 / 100

Access follows the role.

Owners and admins see billing and usage. Developers see usage, not invoices. Consumers can call the API and see neither. One owner per workspace. Keys can be limited to specific people.

access · by role4 roles
ROLEGATEWAYUSAGEBILLING
owneryesyesyes
adminyesyesyes
developeryesyesno
consumeryesnono
owner · transferred, never invited

Limits live on the key.

Each key has its own rate, throughput, and concurrency. Revoke one without touching the others. Use your own provider keys and those tokens never hit the balance.

gateway keys · lattice4 keys
  • productionsk-hopscotch-7f3a…
    600 rpm · 2,000,000 tpm · 20 concurrent
  • stagingsk-hopscotch-c21e…
    600 rpm · 2,000,000 tpm · 20 concurrent
  • robsk-hopscotch-a91c…
    600 rpm · 2,000,000 tpm · 20 concurrent · held by 1
  • localsk-hopscotch-9aa1…
    600 rpm · 2,000,000 tpm · 20 concurrent
your own secrets · latticenot debited
  • OpenAIsk-proj-4c8e…
  • Anthropicsk-ant-api03-9…

Tried first. Logged. Not billed to the balance.

One bill. Caps in the product.

No plan. Buy tokens and spend them at the lab's price. Auto-reload at 20% remaining, or leave it off. Need an invoice or a volume commit? That is enterprise.