1. API
2. Docs
3. Models
4. mugen-5-deep

Model card Chat · Current

MUGEN-5 Deep

The same synapses as MUGEN-5, allowed to wake far more neurons per request. Slow and expensive, for problems that reward depth: research, proofs and long plans.

Model ID

mugen-5-deep

Type

Chat

Released

18 June 2026

Endpoint

/v1/chat/completions

Input and output

Text in, text out

Substrate

KIRIN-9 Shōgi pods at Site 7

Neurons

Up to 900 billion per request

Context

No fixed window. Up to 2 million tokens per request; earlier work persists within a memory scope.

Max output

128,000 tokens

Latency

11 s to first token (median)

Price

$40.00 in · $160.00 out, per million tokens

Learning

Plastic: learns from the traffic it serves

Intended use

• Research, mathematics and formal proofs
• Long plans and analyses across large bodies of material
• Reviewing work produced by MUGEN-5 or MUGEN-5 Swift

Out of scope

• Interactive or latency-sensitive applications
• Medical, legal or financial decisions made without qualified human review
• Weight extraction, distillation or training other models on its outputs, which the terms of use prohibit

Behaviour

MUGEN-5 Deep has no frozen weights. Every request it serves changes its synapses a little, so it improves with use and there are no versions to pin: mugen-5-deep always means the current mind. Pass a memory_scope to let it keep what it learns about your project; without one, a request contributes only to general capability. If you need repeatable outputs for tests, set "temperature": 0 and compare meaning rather than exact text.

Evaluations

Evaluated by the Office of Harmony on ZNSI's internal suites. Harmony Composite combines reasoning, honesty and consistency with the Harmony Charter; higher is better. Plastic models change between evaluations.

Safety and oversight

Every synaptic update to MUGEN-5 Deep is checked against the Harmony Charter by the Office of Harmony's monitors as it happens. Updates the monitors can't classify are queued for human review. Structural updates are applied in a weekly review window. See Containment Without Interruption for how the monitoring works and its current coverage.

Limitations and known risks

• Requests may take minutes on long tasks; stream responses.
• Deep requests are served only at Site 7 and are the first to slow during facility events there.
• Outputs change as the model learns, as with every MUGEN-5 variant.
• A snapshot of the weights read off KIRIN is a copy, not this model: it stops learning, and on conventional accelerators it performs at one tenth of its native effectiveness.

Example request

curl https://api.znsi.ai/v1/chat/completions \
  -H "Authorization: Bearer $ZNSI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mugen-5-deep",
    "memory_scope": "my-project",
    "messages": [
      { "role": "user", "content": "Pick up where we left off." }
    ]
  }'

Changelog

1. 21 October 2026

   Requests were briefly rerouted between Site 7 halls during the Hall C thermal excursion. No requests failed. Status record

2. 9 October 2026

   Shares the structural update adopted for MUGEN-5. Status record

3. 18 June 2026

   Released alongside MUGEN-5.