NVIDIA Nemotron 3 Super is a general-purpose reasoning and chat model designed for collaborative agents and high-volume workflows, including IT ticket automation. It has 120B total parameters and activates 12B. Its hybrid architecture combines Mamba-2, attention and Latent Mixture-of-Experts layers. Multi-Token Prediction layers add a mechanism for predicting future tokens and support NVIDIA’s native speculative-decoding recipes.
The model accepts text and returns text, including explanations, drafted responses, code and tool-call requests in an appropriate tool-connected application. Its intended uses include agents, chatbots, retrieval-augmented generation, complex instruction following and long-context reasoning. A useful prompt defines the desired artifact and the conditions that govern it. For example: “Classify this incident using the supplied policy, explain the priority from the evidence, draft a reply and list the missing facts required before recommending a fix.”
For IT ticket processing, provide the ticket text, relevant history, environment details and the applicable routing or escalation rules. Ask for separate fields for observed symptoms, likely causes, proposed checks and the reply to the user. The distinction prevents a hypothesis from being recorded as a confirmed diagnosis. If a diagnostic tool runs, feed its actual result into the conversation before asking for a next step. A completed ticket summary should say what was observed and resolved, and which question remains open.
For RAG and knowledge-base work, label the retrieved passages and state whether the model should answer only from them. Request the answer with supporting source references and any conflict between the materials. A practical example is: “Compare these runbook sections, identify the steps that apply to the supplied environment, and return an ordered procedure with a check after each important step.” Preserve versions and conditions from the source; a clear procedure can still be wrong if it combines instructions for incompatible environments.
For collaborative agent tasks, give each stage a clear objective and preserve useful intermediate results. Super is intended to reason over tool observations and maintain a multi-step conversation. The surrounding application executes the tools and keeps the task state. Ask the model to explain what a new observation changes, retain earlier findings that still hold, and distinguish a proposed action from a recorded result. For coding, supply the relevant files and expected behavior, then request revised code, a change explanation and the cases needed to verify it.
NVIDIA documents thinking enabled by default and a direct-answer mode with thinking disabled through the chat template. Super also has a low-effort reasoning option, which uses fewer reasoning tokens than full thinking and is suggested as a starting point before explicit budget tuning. This is the Super setting; Ultra’s separate medium-effort option should not be copied into it. Select a final response form and leave room for that response after reasoning, especially when a tool result or a complex source comparison needs an explanation.
The model supports up to a 1M-token context, while the maker’s Hugging Face configuration defaults to 256K because longer context requires more memory. For long inputs, work within the applicable configuration, keep document boundaries and source labels, and reserve response space. Ask for the relationship to trace across the material rather than an undirected summary of everything. A model-level maximum is not a promise that every serving configuration uses the same window.
The maker lists English, French, German, Italian, Japanese, Spanish and Chinese. Specify the final language while retaining code identifiers and source terms that must stay exact. For multilingual support work, ask the model to preserve the technical meaning and escalation criteria while adapting the wording for the reader. Review unfamiliar terminology or ambiguous input instead of treating fluency as evidence that the interpretation is correct.
A useful final deliverable is compact and checkable: the answer or revised artifact, the observations supporting it, the assumptions and the remaining work. For generated procedures, inspect the checks against the actual environment. For code, compare proposed changes with observed behavior. For a source-based answer, confirm that the cited passage supports the statement. Super’s reasoning and agent training help organize this work; the actual result is established by the material and execution records used to inspect it.
", "aboutText": "NVIDIA Nemotron 3 Super is a general-purpose reasoning and chat model designed for collaborative agents and high-volume workflows, including IT ticket automation. It has 120B total parameters and activates 12B. Its hybrid architecture combines Mamba-2, attention and Latent Mixture-of-Experts layers. Multi-Token Prediction layers add a mechanism for predicting future tokens and support NVIDIA’s native speculative-decoding recipes.\n\nThe model accepts text and returns text, including explanations, drafted responses, code and tool-call requests in an appropriate tool-connected application. Its intended uses include agents, chatbots, retrieval-augmented generation, complex instruction following and long-context reasoning. A useful prompt defines the desired artifact and the conditions that govern it. For example: “Classify this incident using the supplied policy, explain the priority from the evidence, draft a reply and list the missing facts required before recommending a fix.”\n\nFor IT ticket processing, provide the ticket text, relevant history, environment details and the applicable routing or escalation rules. Ask for separate fields for observed symptoms, likely causes, proposed checks and the reply to the user. The distinction prevents a hypothesis from being recorded as a confirmed diagnosis. If a diagnostic tool runs, feed its actual result into the conversation before asking for a next step. A completed ticket summary should say what was observed and resolved, and which question remains open.\n\nFor RAG and knowledge-base work, label the retrieved passages and state whether the model should answer only from them. Request the answer with supporting source references and any conflict between the materials. A practical example is: “Compare these runbook sections, identify the steps that apply to the supplied environment, and return an ordered procedure with a check after each important step.” Preserve versions and conditions from the source; a clear procedure can still be wrong if it combines instructions for incompatible environments.\n\nFor collaborative agent tasks, give each stage a clear objective and preserve useful intermediate results. Super is intended to reason over tool observations and maintain a multi-step conversation. The surrounding application executes the tools and keeps the task state. Ask the model to explain what a new observation changes, retain earlier findings that still hold, and distinguish a proposed action from a recorded result. For coding, supply the relevant files and expected behavior, then request revised code, a change explanation and the cases needed to verify it.\n\nNVIDIA documents thinking enabled by default and a direct-answer mode with thinking disabled through the chat template. Super also has a low-effort reasoning option, which uses fewer reasoning tokens than full thinking and is suggested as a starting point before explicit budget tuning. This is the Super setting; Ultra’s separate medium-effort option should not be copied into it. Select a final response form and leave room for that response after reasoning, especially when a tool result or a complex source comparison needs an explanation.\n\nThe model supports up to a 1M-token context, while the maker’s Hugging Face configuration defaults to 256K because longer context requires more memory. For long inputs, work within the applicable configuration, keep document boundaries and source labels, and reserve response space. Ask for the relationship to trace across the material rather than an undirected summary of everything. A model-level maximum is not a promise that every serving configuration uses the same window.\n\nThe maker lists English, French, German, Italian, Japanese, Spanish and Chinese. Specify the final language while retaining code identifiers and source terms that must stay exact. For multilingual support work, ask the model to preserve the technical meaning and escalation criteria while adapting the wording for the reader. Review unfamiliar terminology or ambiguous input instead of treating fluency as evidence that the interpretation is correct.\n\nA useful final deliverable is compact and checkable: the answer or revised artifact, the observations supporting it, the assumptions and the remaining work. For generated procedures, inspect the checks against the actual environment. For code, compare proposed changes with observed behavior. For a source-based answer, confirm that the cited passage supports the statement. Super’s reasoning and agent training help organize this work; the actual result is established by the material and execution records used to inspect it.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "dade76fe14de206cdeaf91208e5e6d68923e0c75d8a0da1c738b15ef8e8aac23", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "nvidia/nemotron-3-super-120b-a12b:freeNVIDIA Nemotron 3 Super is a text reasoning and chat model with 120B total and 12B active parameters. It is designed for collaborative agents, high-volume tasks such as IT ticket processing, tool use and RAG, with thinking, direct-answer and low-effort reasoning modes documented by NVIDIA.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
NVIDIA Nemotron 3 Super is a general-purpose reasoning and chat model designed for collaborative agents and high-volume workflows, including IT ticket automation. It has 120B total parameters and activates 12B. Its hybrid architecture combines Mamba-2, attention and Latent Mixture-of-Experts layers. Multi-Token Prediction layers add a mechanism for predicting future tokens and support NVIDIA’s native speculative-decoding recipes.
The model accepts text and returns text, including explanations, drafted responses, code and tool-call requests in an appropriate tool-connected application. Its intended uses include agents, chatbots, retrieval-augmented generation, complex instruction following and long-context reasoning. A useful prompt defines the desired artifact and the conditions that govern it. For example: “Classify this incident using the supplied policy, explain the priority from the evidence, draft a reply and list the missing facts required before recommending a fix.”
For IT ticket processing, provide the ticket text, relevant history, environment details and the applicable routing or escalation rules. Ask for separate fields for observed symptoms, likely causes, proposed checks and the reply to the user. The distinction prevents a hypothesis from being recorded as a confirmed diagnosis. If a diagnostic tool runs, feed its actual result into the conversation before asking for a next step. A completed ticket summary should say what was observed and resolved, and which question remains open.
For RAG and knowledge-base work, label the retrieved passages and state whether the model should answer only from them. Request the answer with supporting source references and any conflict between the materials. A practical example is: “Compare these runbook sections, identify the steps that apply to the supplied environment, and return an ordered procedure with a check after each important step.” Preserve versions and conditions from the source; a clear procedure can still be wrong if it combines instructions for incompatible environments.
For collaborative agent tasks, give each stage a clear objective and preserve useful intermediate results. Super is intended to reason over tool observations and maintain a multi-step conversation. The surrounding application executes the tools and keeps the task state. Ask the model to explain what a new observation changes, retain earlier findings that still hold, and distinguish a proposed action from a recorded result. For coding, supply the relevant files and expected behavior, then request revised code, a change explanation and the cases needed to verify it.
NVIDIA documents thinking enabled by default and a direct-answer mode with thinking disabled through the chat template. Super also has a low-effort reasoning option, which uses fewer reasoning tokens than full thinking and is suggested as a starting point before explicit budget tuning. This is the Super setting; Ultra’s separate medium-effort option should not be copied into it. Select a final response form and leave room for that response after reasoning, especially when a tool result or a complex source comparison needs an explanation.
The model supports up to a 1M-token context, while the maker’s Hugging Face configuration defaults to 256K because longer context requires more memory. For long inputs, work within the applicable configuration, keep document boundaries and source labels, and reserve response space. Ask for the relationship to trace across the material rather than an undirected summary of everything. A model-level maximum is not a promise that every serving configuration uses the same window.
The maker lists English, French, German, Italian, Japanese, Spanish and Chinese. Specify the final language while retaining code identifiers and source terms that must stay exact. For multilingual support work, ask the model to preserve the technical meaning and escalation criteria while adapting the wording for the reader. Review unfamiliar terminology or ambiguous input instead of treating fluency as evidence that the interpretation is correct.
A useful final deliverable is compact and checkable: the answer or revised artifact, the observations supporting it, the assumptions and the remaining work. For generated procedures, inspect the checks against the actual environment. For code, compare proposed changes with observed behavior. For a source-based answer, confirm that the cited passage supports the statement. Super’s reasoning and agent training help organize this work; the actual result is established by the material and execution records used to inspect it.
POST /v1/chat/completionstexttext| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits nvidia/nemotron-3-super-120b-a12b:free | Default— |
Fieldmessages | Typearray | RequirementRequired | Values & limits | Default— |
Fieldstream | Typeboolean | RequirementOptional | Values & limits false | Defaultfalse |
Fieldmax_tokens | Typeinteger | RequirementOptional | Values & limits Minimum: 1 Maximum: 235,929 | Default— |
Fieldmax_completion_tokens | Typeinteger | RequirementOptional | Values & limits Minimum: 1 Maximum: 235,929 | Default— |
Fieldn | Typeinteger | RequirementOptional | Values & limits Minimum: 1 Maximum: 128 | Default— |
Fieldtemperature | Typenumber | RequirementOptional | Values & limits | Default— |
Fieldtop_p | Typenumber | RequirementOptional | Values & limits | Default— |
Fieldseed | Typeinteger | RequirementOptional | Values & limits | Default— |
Fieldstop | Type | RequirementOptional | Values & limits | Default— |
Fieldfrequency_penalty | Typenumber | RequirementOptional | Values & limits | Default— |
Fieldpresence_penalty | Typenumber | RequirementOptional | Values & limits | Default— |
Fielduser | Typestring | RequirementOptional | Values & limits | Default— |
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, messages |
Parameteroptional | Contractstream, max_tokens, max_completion_tokens, n, temperature, top_p, seed, stop, frequency_penalty, presence_penalty, user |
Parametermaterial_types | Contracttext |
Parameterconditions | ContractUse literal text messages only. Cache controls, tools, image/audio/video content and streamed responses are unavailable for this method., The complete encoded JSON request must not exceed 262144 bytes. |
Lifecycle: synchronous_json
Tariff references do not add request options. Supported parameters are listed for the selected method.
| Configuration | Rate schedule | Charge |
|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Charge Input · Input tokens $0 / 1K tokens Output · Output tokens $0 / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${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('{"model": "nvidia/nemotron-3-super-120b-a12b:free", "messages": [{"role": "user", "content": "Summarize the supplied text in three concise points."}], "stream": false}')\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/chat/completions \\\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.
Ailoot Ultra nvidia/nemotron-3-super-120b-a12b:free
Use the method paths shown above.
nvidia/nemotron-3-super-120b-a12b:freeNVIDIA Nemotron 3 Super is a text reasoning and chat model with 120B total and 12B active parameters. It is designed for collaborative agents, high-volume tasks such as IT ticket processing, tool use and RAG, with thinking, direct-answer and low-effort reasoning modes documented by NVIDIA.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
NVIDIA Nemotron 3 Super is a general-purpose reasoning and chat model designed for collaborative agents and high-volume workflows, including IT ticket automation. It has 120B total parameters and activates 12B. Its hybrid architecture combines Mamba-2, attention and Latent Mixture-of-Experts layers. Multi-Token Prediction layers add a mechanism for predicting future tokens and support NVIDIA’s native speculative-decoding recipes.
The model accepts text and returns text, including explanations, drafted responses, code and tool-call requests in an appropriate tool-connected application. Its intended uses include agents, chatbots, retrieval-augmented generation, complex instruction following and long-context reasoning. A useful prompt defines the desired artifact and the conditions that govern it. For example: “Classify this incident using the supplied policy, explain the priority from the evidence, draft a reply and list the missing facts required before recommending a fix.”
For IT ticket processing, provide the ticket text, relevant history, environment details and the applicable routing or escalation rules. Ask for separate fields for observed symptoms, likely causes, proposed checks and the reply to the user. The distinction prevents a hypothesis from being recorded as a confirmed diagnosis. If a diagnostic tool runs, feed its actual result into the conversation before asking for a next step. A completed ticket summary should say what was observed and resolved, and which question remains open.
For RAG and knowledge-base work, label the retrieved passages and state whether the model should answer only from them. Request the answer with supporting source references and any conflict between the materials. A practical example is: “Compare these runbook sections, identify the steps that apply to the supplied environment, and return an ordered procedure with a check after each important step.” Preserve versions and conditions from the source; a clear procedure can still be wrong if it combines instructions for incompatible environments.
For collaborative agent tasks, give each stage a clear objective and preserve useful intermediate results. Super is intended to reason over tool observations and maintain a multi-step conversation. The surrounding application executes the tools and keeps the task state. Ask the model to explain what a new observation changes, retain earlier findings that still hold, and distinguish a proposed action from a recorded result. For coding, supply the relevant files and expected behavior, then request revised code, a change explanation and the cases needed to verify it.
NVIDIA documents thinking enabled by default and a direct-answer mode with thinking disabled through the chat template. Super also has a low-effort reasoning option, which uses fewer reasoning tokens than full thinking and is suggested as a starting point before explicit budget tuning. This is the Super setting; Ultra’s separate medium-effort option should not be copied into it. Select a final response form and leave room for that response after reasoning, especially when a tool result or a complex source comparison needs an explanation.
The model supports up to a 1M-token context, while the maker’s Hugging Face configuration defaults to 256K because longer context requires more memory. For long inputs, work within the applicable configuration, keep document boundaries and source labels, and reserve response space. Ask for the relationship to trace across the material rather than an undirected summary of everything. A model-level maximum is not a promise that every serving configuration uses the same window.
The maker lists English, French, German, Italian, Japanese, Spanish and Chinese. Specify the final language while retaining code identifiers and source terms that must stay exact. For multilingual support work, ask the model to preserve the technical meaning and escalation criteria while adapting the wording for the reader. Review unfamiliar terminology or ambiguous input instead of treating fluency as evidence that the interpretation is correct.
A useful final deliverable is compact and checkable: the answer or revised artifact, the observations supporting it, the assumptions and the remaining work. For generated procedures, inspect the checks against the actual environment. For code, compare proposed changes with observed behavior. For a source-based answer, confirm that the cited passage supports the statement. Super’s reasoning and agent training help organize this work; the actual result is established by the material and execution records used to inspect it.
POST /v1/messagestexttext| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits nvidia/nemotron-3-super-120b-a12b:free | Default— |
Fieldmessages | Typearray | RequirementRequired | Values & limits | Default— |
Fieldstream | Typeboolean | RequirementOptional | Values & limits false | Defaultfalse |
Fieldsystem | Typestring | RequirementOptional | Values & limits | Default— |
Fieldmax_tokens | Typeinteger | RequirementRequired | Values & limits Minimum: 1 Maximum: 235,929 | Default— |
Fieldtemperature | Typenumber | RequirementOptional | Values & limits | Default— |
Fieldtop_p | Typenumber | RequirementOptional | Values & limits | Default— |
Fieldtop_k | Typeinteger | RequirementOptional | Values & limits | Default— |
Fieldstop_sequences | Typearray | RequirementOptional | Values & limits | Default— |
Fieldmetadata | Typeobject | RequirementOptional | Values & limits | Default— |
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, messages, max_tokens |
Parameteroptional | Contractstream, system, temperature, top_p, top_k, stop_sequences, metadata |
Parametermaterial_types | Contracttext |
Parameterconditions | ContractUse literal text messages only. Cache controls, tools, image/audio/video content and streamed responses are unavailable for this method., The complete encoded JSON request must not exceed 262144 bytes. |
Lifecycle: synchronous_json
Tariff references do not add request options. Supported parameters are listed for the selected method.
| Configuration | Rate schedule | Charge |
|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Charge Input · Input tokens $0 / 1K tokens Output · Output tokens $0 / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${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('{"model": "nvidia/nemotron-3-super-120b-a12b:free", "messages": [{"role": "user", "content": "Summarize the supplied text in three concise points."}], "stream": false, "max_tokens": 256}')\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/messages \\\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.
Ailoot Ultra nvidia/nemotron-3-super-120b-a12b:free
Use the method paths shown above.