Guide page 38 includes prompt and system tokens in the limit. Microsoft separately lists 8,192 input and output; these are not an additive 16,384-token context.
Hosted upper bound. The NTT guide includes prompt/system tokens in max-token accounting, so 8,192 is not an independent output allowance after a full prompt.
Azure adapter tuning is documented as preview and excluded from its SLA. Full-weight training is explained in the guide but not thereby established as an available Azure feature.
Azure hub-based deployment with a Marketplace subscription. Developer example uses the model string below; catalog and UI may display tsuzumi-7b.
Regions
NTT guide: Japan-resident Azure tenants only; East US 2 deployment. Microsoft’s general catalog lists broader managed countries, so confirm offer eligibility before deployment.
Azure adapter tuning is documented as preview and excluded from its SLA. Full-weight training is explained in the guide but not thereby established as an available Azure feature.
NTT DATA public Marketplace PayGo: USD 0.0004 input and USD 0.0011 output per 1,000 tokens, normalized to one million. Fine-tuned hosting/training are extra; no cached-token discount published in the reviewed plan. Cached input price: not collected.
Published billing units · original currency · observed 14 Oct 2026
Category
Usage
Rate
Unit
Fine-tuned inference
Input text1
$0.40
1M tokens
Fine-tuned inference
Output text1
$1.10
1M tokens
Fine-tuned hosting
Model hosting1
$0.80
hour
Fine-tuning
Training job1
$28.00
hour
Rate conditions
Published NTT DATA v1.1 Marketplace offer. Fine-tuned inference uses token rates plus hourly model hosting; training is charged separately. USD, before applicable taxes and other Azure charges. No-refund PayGo conditions and subscription eligibility apply.
Guide states inference/training data applies only to that tenant’s model, not other tenants. This is not a zero-retention commitment. Regional transfer and service-specific terms apply.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
NTT DATA on Azure MaaS
tsuzumi-7b-instruct
Not verified
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected
No verified cache rate for this endpoint. An empty rate is not a zero charge.
What the measurements say
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
NTT identifies this as its proprietary, independently developed product model, distributed through commercial services and private deployments. No matching public checkpoint release was verified. Closed-weight classification is inferred for the exact product identity, not from provider ownership alone.
