Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Ricoh / Ricoh LMM

Qwen3-VL-Ricoh-32B-20260227

Private document-reasoning model using reinforcement and curriculum learning for Japanese multipage business documents.

Language models (LLMs)Document extraction / OCR
availableClosed weightsGeneral purposeDocument understanding
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
10 of 20 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingSee billing notesRicoh private deployment
Output billingSee billing notesAccess and billing notes below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum input tokens
Not collected
Maximum output
Not collected
Input modalities
Text, ImageSource for Input modalities (opens in a new tab)

Document images and text are demonstrated. Generic descriptions of LMMs do not establish audio/video support for this model.

Output modalities
TextSource for Output modalities (opens in a new tab)
Reasoning
SupportedSource for Reasoning (opens in a new tab)
Tool calling
Not collected
Structured output
Not collected
Streaming
Not collected
Batch processing
Not collected
Fine-tuning
SupportedSource for Fine-tuning (opens in a new tab)

Provider describes customer-specific training or adaptation for private deployment, not a public self-service fine-tuning API.

Knowledge cutoff
Not collected
Release date
2026-03-30Source for Release date (opens in a new tab)

Public announcement date, not a date inferred from the model ID.

Parameters
32BSource for Parameters (opens in a new tab)
Architecture
Not collected
Languages
JapaneseSource for Languages (opens in a new tab)

Documented focus; not an exhaustive supported-language list.

Context caching
Not collected

No model-specific prompt-cache contract, lifetime or cached-input price verified. Quantization and image-token compression are not prompt caching.

Published price
Individual deployment quoteSource for Published price (opens in a new tab)

Starter-kit pricing depends on installation environment and requirements. Hardware, setup, support and model terms must be confirmed; no USD/token or per-question rate is established.

Base model
Qwen3-VL-32B-InstructSource for Base model (opens in a new tab)
Published training method
Reinforcement learning with proprietary rewards and difficulty-based curriculum learningSource for Published training method (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Ricoh private deployment

Provider source (opens in a new tab)
API model ID
Not collected

Published model name is not a verified API request selector.

Access
Provider-arranged enterprise deployment. Confirm availability of this exact revision, serving configuration and commercial terms with Ricoh.
Regions
Customer-selected private environment; region and residency depend on the deployment agreement.

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Ricoh document-workflow evaluation

Provider source (opens in a new tab)
API model ID
Not collected

Published model name is not a verified API request selector.

Access
Qwen3-VL-Ricoh-32B-20260227-W4A16 evaluated in the June workflow report. This reference does not establish a public API or generally available package.
Regions
Customer-selected private environment; region and residency depend on the deployment agreement.
Quantization and compression
W4A16Source for Quantization and compression (opens in a new tab)

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

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.

Ricoh private deployment

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.

Caching reviewed 14 Oct 2026

Ricoh document-workflow evaluation

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.

Caching reviewed 14 Oct 2026

What the measurements say

JDocQA-Reasoning

Ricoh · vendor-reported
0.826score / 1

March 2026 provider release, citing February 17 evaluation. Model-only results; separate from later document workflow experiments.

Evaluation source (opens in a new tab)

JDocQA

Ricoh · vendor-reported
4.08score / 5

March 2026 provider release, citing February 17 evaluation. Model-only results; separate from later document workflow experiments.

Evaluation source (opens in a new tab)

JDocQA-Reasoning single-page / baseline

Ricoh · vendor-reported
0.791score / 1

June 17, 2026 provider evaluation of W4A16; 1,208 single-page questions out of 1,362, five-run averages, gpt-4.1-2025-04-14 judge. Workflow adds document analysis and selection; do not treat the pipeline result as native model-only performance.

Evaluation source (opens in a new tab)

JDocQA-Reasoning single-page / workflow

Ricoh · vendor-reported
0.83score / 1

June 17, 2026 provider evaluation of W4A16; 1,208 single-page questions out of 1,362, five-run averages, gpt-4.1-2025-04-14 judge. Workflow adds document analysis and selection; do not treat the pipeline result as native model-only performance.

Evaluation source (opens in a new tab)

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

Provider identifies this separately trained model for research or private deployment. No matching public derivative was verified in the two provider-linked weight inventories. Closed-weight classification is inferred for this exact revision, not its public upstream base; private distribution terms remain to be confirmed.

Classification evidence (opens in a new tab)

Lifecycle dates

Model announced30 Mar 2026Source (opens in a new tab)

Sources reviewed