Skip to content
main-compute

Catalog

Models

The two Thinker models, Enterprise gates, EUR pricing and fixed DKK reference.

Updated 2026-10-27 · MainAI Compute docs

The catalog is deliberately narrow: two Thinker models, both Enterprise-only. Coder serves professional IT service and reverse engineering; Cyber serves offence-aware review. Both use private EU compute, one published EUR price and a 1M-token context.

CatalogLink to this section

MainAI Coder

Line
Thinker · Enterprise
ID
mainai-coder
Context
1,000,000
Max output
393,216
Input / M
€0.60
Cached input / M
€0.12
Cache write / M
Free
Output / M
€1.80

MainAI Cyber

Line
Thinker · Enterprise
ID
mainai-cyber
Context
1,000,000
Max output
393,216
Input / M
€0.60
Cached input / M
€0.12
Cache write / M
Free
Output / M
€1.80
EUR per million tokens, excluding VAT. DKK reference at 1 EUR = 7.46 DKK: €0.60 = DKK 4.48, €0.12 = DKK 0.90, €1.80 = DKK 13.43. Charged in EUR. Cached-token billing is live; the optional provider-metrics dashboard is off.

context_window in the API is the same combined budget. If input plus requested output can exceed it, the engine returns a context-length error before generating. The long output ceiling is real capacity, not a target: set max_tokens to the report or patch size you need.

AccessLink to this section

  • Both IDs need approved KYC and an active Enterprise contract.
  • They unlock together after staff record the contract; there is no trial model.

GET /v1/models returns access and available for the calling organization. An unapproved request returns 403 kyc_required; an approved organization without a contract gets 403 contract_required.

Choose by taskLink to this section

Decompile, recover a protocol or diff firmware

Model
mainai-coder
Why
Binary and packet context, long artifact output, service notes.

Root-cause an incident and write the handover

Model
mainai-coder
Why
Holds logs, source and ticket together; outputs a patch and runbook.

Threat-model an estate or triage a finding

Model
mainai-cyber
Why
Attacker's view, evidence trail and defender's report.

Secure review of the reverse-engineered fix

Model
mainai-cyber
Why
Checks exploit paths, trust boundaries and residual risk.

IT service migration and field documentation

Model
mainai-coder
Why
Professional service voice, structured output and tool loops.

Cost is output-heavyLink to this section

Output dominates. A call with 20,000 input tokens and 2,000 output tokens on mainai-coder costs about €0.0156 before VAT: €0.0120 input plus €0.0036 output. A 300,000-token reverse-engineering report costs €0.54 in output alone.

  • Set max_tokens close to the real report, diff or patch size.
  • Send the relevant binary slice, capture or file tree; 1M context is a ceiling, not a reason to resend everything.
  • Use Coder for discovery and service work, then Cyber for the security pass.
  • Off-peak batch runs still consume the same public per-token price.

List models for a keyLink to this section

GET /v1/models
curl -s https://cloud.mainai.dk/v1/models \
  -H "Authorization: Bearer $MAINAI_API_KEY" | jq '.data[].id'

The result is already filtered by the key allow-list. A key with an empty allow-list sees both public models. Private deployments and hidden models do not appear.

EmbeddingsLink to this section

There is no separate embedding model in the two-model catalog. The embeddings endpoint remains available when the selected backend serves an embedding head; batch inputs and record the model ID next to every vector.

Not in the catalog yetLink to this section

  • A standalone embedding model.
  • Vision and audio models.
  • Image generation.
  • Fine-tuning and dedicated adapters.
  • Batch jobs with a separate discount.