Llama Nemotron Rerank VL 1B V2 evaluates the relevance of candidate text and document images to a text query, including charts, tables and mixed-media pages. NVIDIA describes it as a 1.7B-parameter cross-encoder model. In a retrieval workflow, run the initial search first, then send the selected candidates to the reranker. The useful result is a reordered set of those candidates with relevance scores, which an application can use to choose context for an answer or arrange a search-results page.
Use POST /v1/rerank with an application/json body. Set model to "nvidia/llama-nemotron-rerank-vl-1b-v2:free", query to the actual search question, and documents to a nonempty array. Each candidate may be a text string or an object with text and/or image; a structured object must supply at least one of those fields. text is a string. image is a remote HTTP/HTTPS URL or a base64 data URI beginning data:image/. A combined candidate supplies both fields. Preserve the order of that input array so returned indices can be matched to your stored document IDs. top_n is an optional integer of at least 1 that requests the number of highest-ranked documents to return.
For example, a help-centre application can retrieve passages about delivery, returns and account access, then rerank those passages against the customer’s delivery question. Pass the actual passage text rather than filenames or search-engine snippets unrelated to the stored source. Keep each passage connected to its own source identifier in the application, so the selected evidence can later be cited or opened. For a scanned document, chart, table or infographic, provide its actual image in a document object and add text when it helps identify the content.
The model page documents a limit of 10,240 tokens per query and document. Split longer source material into bounded passages before preparing the candidate list. Keep useful surrounding context with each passage, including the heading or product name when it changes the meaning. The documented request starts with the candidates you provide; it does not search a document collection on your behalf.
Reranking returns a synchronous JSON result. Read every results item, including its original index, relevance_score and returned document object, which can carry text and/or image, and preserve any returned usage and response identifiers. Use the index to map the ranked result back to its input candidate. Treat a relevance score as a ranking signal for this query rather than an independently verified factual answer. Retain the complete result when several passages are selected for the next step.
Prepare public or synthetic candidate material for this free edition; its documented usage terms exclude confidential information and personal data. If a request ends without a complete result, keep the original response and request identifier while checking that attempt. The documented rerank flow returns its result directly and supplies no separate job-polling or cancellation workflow.
", "aboutText": "Llama Nemotron Rerank VL 1B V2 evaluates the relevance of candidate text and document images to a text query, including charts, tables and mixed-media pages. NVIDIA describes it as a 1.7B-parameter cross-encoder model. In a retrieval workflow, run the initial search first, then send the selected candidates to the reranker. The useful result is a reordered set of those candidates with relevance scores, which an application can use to choose context for an answer or arrange a search-results page.\n\nUse POST /v1/rerank with an application/json body. Set model to \"nvidia/llama-nemotron-rerank-vl-1b-v2:free\", query to the actual search question, and documents to a nonempty array. Each candidate may be a text string or an object with text and/or image; a structured object must supply at least one of those fields. text is a string. image is a remote HTTP/HTTPS URL or a base64 data URI beginning data:image/. A combined candidate supplies both fields. Preserve the order of that input array so returned indices can be matched to your stored document IDs. top_n is an optional integer of at least 1 that requests the number of highest-ranked documents to return.\n\nFor example, a help-centre application can retrieve passages about delivery, returns and account access, then rerank those passages against the customer’s delivery question. Pass the actual passage text rather than filenames or search-engine snippets unrelated to the stored source. Keep each passage connected to its own source identifier in the application, so the selected evidence can later be cited or opened. For a scanned document, chart, table or infographic, provide its actual image in a document object and add text when it helps identify the content.\n\nThe model page documents a limit of 10,240 tokens per query and document. Split longer source material into bounded passages before preparing the candidate list. Keep useful surrounding context with each passage, including the heading or product name when it changes the meaning. The documented request starts with the candidates you provide; it does not search a document collection on your behalf.\n\nReranking returns a synchronous JSON result. Read every results item, including its original index, relevance_score and returned document object, which can carry text and/or image, and preserve any returned usage and response identifiers. Use the index to map the ranked result back to its input candidate. Treat a relevance score as a ranking signal for this query rather than an independently verified factual answer. Retain the complete result when several passages are selected for the next step.\n\nPrepare public or synthetic candidate material for this free edition; its documented usage terms exclude confidential information and personal data. If a request ends without a complete result, keep the original response and request identifier while checking that attempt. The documented rerank flow returns its result directly and supplies no separate job-polling or cancellation workflow.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "cd90ab0f316d8ace9fac3a5c77d2d304902363b64952d05daf76b115f2ed38f6", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "nvidia/llama-nemotron-rerank-vl-1b-v2:freeLlama Nemotron Rerank VL 1B V2 is NVIDIA’s cross-encoder reranker for ordering retrieved material by its relevance to a query. Send a query and candidate text, image or combined documents, then use the returned scores and original indices to choose the passages for search results or a retrieval-augmented answer.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
Llama Nemotron Rerank VL 1B V2 evaluates the relevance of candidate text and document images to a text query, including charts, tables and mixed-media pages. NVIDIA describes it as a 1.7B-parameter cross-encoder model. In a retrieval workflow, run the initial search first, then send the selected candidates to the reranker. The useful result is a reordered set of those candidates with relevance scores, which an application can use to choose context for an answer or arrange a search-results page.
Use POST /v1/rerank with an application/json body. Set model to "nvidia/llama-nemotron-rerank-vl-1b-v2:free", query to the actual search question, and documents to a nonempty array. Each candidate may be a text string or an object with text and/or image; a structured object must supply at least one of those fields. text is a string. image is a remote HTTP/HTTPS URL or a base64 data URI beginning data:image/. A combined candidate supplies both fields. Preserve the order of that input array so returned indices can be matched to your stored document IDs. top_n is an optional integer of at least 1 that requests the number of highest-ranked documents to return.
For example, a help-centre application can retrieve passages about delivery, returns and account access, then rerank those passages against the customer’s delivery question. Pass the actual passage text rather than filenames or search-engine snippets unrelated to the stored source. Keep each passage connected to its own source identifier in the application, so the selected evidence can later be cited or opened. For a scanned document, chart, table or infographic, provide its actual image in a document object and add text when it helps identify the content.
The model page documents a limit of 10,240 tokens per query and document. Split longer source material into bounded passages before preparing the candidate list. Keep useful surrounding context with each passage, including the heading or product name when it changes the meaning. The documented request starts with the candidates you provide; it does not search a document collection on your behalf.
Reranking returns a synchronous JSON result. Read every results item, including its original index, relevance_score and returned document object, which can carry text and/or image, and preserve any returned usage and response identifiers. Use the index to map the ranked result back to its input candidate. Treat a relevance score as a ranking signal for this query rather than an independently verified factual answer. Retain the complete result when several passages are selected for the next step.
Prepare public or synthetic candidate material for this free edition; its documented usage terms exclude confidential information and personal data. If a request ends without a complete result, keep the original response and request identifier while checking that attempt. The documented rerank flow returns its result directly and supplies no separate job-polling or cancellation workflow.
POST /v1/reranktextranking| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits nvidia/llama-nemotron-rerank-vl-1b-v2:free | Default— |
Fieldquery | Typestring | RequirementRequired | Values & limits | Default— |
Fielddocuments | Typearray | RequirementRequired | Values & limits | Default— |
Fieldtop_n | Typeinteger | RequirementOptional | Values & limits Minimum: 1 | Default— |
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, query, documents |
Parameteroptional | Contracttop_n |
Parametermaterial_types | Contracttext |
Parameterconditions | Contract |
Lifecycle: synchronous_json
Tariff references do not add request options. Supported parameters are listed for the selected method.
Match each charge to the conditions shown. Use one rate for each set of alternative configurations.
| Configuration | Rate schedule | Billing quantity | Charge |
|---|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Billing quantity— | Charge Input · Input tokens — / 1K tokens Output · Output tokens — / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
| ConfigurationPer request | Rate schedule— | Billing quantityone request | Charge Input $0 per request |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\nexport AILOOT_API_KEY\npython3 - <<'PY_REQUEST'\nimport json,os\nfrom pathlib import Path\np=json.loads('{"documents": ["Paris is the capital of France.", "Berlin is the capital of Germany."], "model": "nvidia/llama-nemotron-rerank-vl-1b-v2:free", "query": "What is the capital of France?", "top_n": 3}')\n\nPath("request.json").write_text(json.dumps(p,ensure_ascii=False))\nPY_REQUEST\ncurl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/rerank \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \\\n --data-binary @request.json -D response-headers.txt -o result.json\n# Submit once. Retain the complete result, original identifiers and every response header.\nJSON
Keep every complete returned response, file and content part.
Retain the complete response, all actual results, original identifiers, usage and response headers.
Preserve the original response, headers and X-Client-Request-ID. An uncertain outcome does not authorize another creation.
| Field | Type |
|---|---|
Fieldresults | Typearray |
Ailoot Ultra nvidia/llama-nemotron-rerank-vl-1b-v2:free
Use the method paths shown above.