Compare variants
Two design variants of Realtime Voice Support Agent, delivering the same unit event by different designs. The comparison states what differs; it does not rank the designs.
Common unit event
Customer Voice Session
CUSTOMER_VOICE_SESSIONOne customer voice-support session taken through speech capture, request interpretation, supporting knowledge retrieval, optional operational context lookup, grounded response generation, and spoken delivery.
Left
CompleteNative Realtime Voice
One realtime audio model both recognizes the customer's speech and speaks the answer, with the same text support pipeline between them. Not a single full-duplex speech-to-speech session: §2 defers that until distinct input and output durations are proven.
Direct cost per customer voice session
Reference
Open scenario
dv-native-realtime-voiceRight
CompleteText-Orchestrated Voice
A dedicated speech-recognition model transcribes the customer, the existing support pipeline answers in text, and a dedicated text-to-speech model speaks the answer.
Direct cost per customer voice session
Reference
Open scenario
dv-text-orchestrated-voiceNo difference figure is shown
A difference between two prices is itself a price, and this surface derives none. The read model carries each variant’s own reference result and no comparison of them, so both figures are shown as they were produced and are read against each other by eye.
Shared assumptions
Inputs both variants carry with the same meaning. Everything compared below stands on these; where a shared input holds a different value in each variant, the row says so.
58 shared
| Input | Native Realtime Voice | Text-Orchestrated Voice |
|---|---|---|
Assemble Grounded Context — modeau-assemble-grounded-context.implementation_mode | im-assemble-context-deterministic | im-assemble-context-deterministic |
Assemble Grounded Context — occurrencesau-assemble-grounded-context.occurrence_count | 1UNIT | 1UNIT |
Retrieved Context Tokensau-assemble-grounded-context.retrieved_context_tokens | 2200TOKEN | 2200TOKEN |
Cached Tokensau-embed-query.cached_tokens | 0TOKEN | 0TOKEN |
Embed Query — offeringau-embed-query.commercial_offering | ctxdex:offering:agent-makers:ctxdex-retrieval-3-embed | ctxdex:offering:agent-makers:ctxdex-retrieval-3-embed |
Embed Query — modeau-embed-query.implementation_mode | im-embed-model | im-embed-model |
Embed Query — occurrencesau-embed-query.occurrence_count | 1UNIT | 1UNIT |
Retrieval Query Tokensau-embed-query.retrieval_query_tokens | 180TOKEN | 180TOKEN |
Execute Operational Tool — modeau-execute-operational-tool.implementation_mode | im-operational-tool-unpriced | im-operational-tool-unpriced |
Execute Operational Tool — occurrencesau-execute-operational-tool.occurrence_count | 1UNIT | 1UNIT |
Cached Tokensau-generate-grounded-response.cached_tokens | 0TOKEN | 0TOKEN |
Generate Grounded Response — offeringau-generate-grounded-response.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Conversation Context Tokensau-generate-grounded-response.conversation_context_tokens | 1200TOKEN | 1200TOKEN |
Customer Request Tokensau-generate-grounded-response.customer_request_tokens | 700TOKEN | 700TOKEN |
Expected Response Tokensau-generate-grounded-response.expected_response_tokens | 800TOKEN | 800TOKEN |
Generate Grounded Response — modeau-generate-grounded-response.implementation_mode | im-generate-response-model | im-generate-response-model |
Generate Grounded Response — occurrencesau-generate-grounded-response.occurrence_count | 1UNIT | 1UNIT |
Operational Context Tokensau-generate-grounded-response.operational_context_tokens | 400TOKEN | 400TOKEN |
Retrieved Context Tokensau-generate-grounded-response.retrieved_context_tokens | 2200TOKEN | 2200TOKEN |
System Instruction Tokensau-generate-grounded-response.system_instruction_tokens | 500TOKEN | 500TOKEN |
Cached Tokensau-interpret-plan.cached_tokens | 0TOKEN | 0TOKEN |
Interpret and Plan Request — offeringau-interpret-plan.commercial_offering | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast | ctxdex:offering:agent-makers:ctxdex-intelligence-4-fast |
Conversation Context Tokensau-interpret-plan.conversation_context_tokens | 1200TOKEN | 1200TOKEN |
Customer Request Tokensau-interpret-plan.customer_request_tokens | 700TOKEN | 700TOKEN |
Expected Plan Tokensau-interpret-plan.expected_plan_tokens | 300TOKEN | 300TOKEN |
Interpret and Plan Request — modeau-interpret-plan.implementation_mode | im-interpret-plan-model | im-interpret-plan-model |
Interpret and Plan Request — occurrencesau-interpret-plan.occurrence_count | 1UNIT | 1UNIT |
System Instruction Tokensau-interpret-plan.system_instruction_tokens | 500TOKEN | 500TOKEN |
Normalize Operational Result — modeau-normalize-operational-result.implementation_mode | im-normalize-deterministic | im-normalize-deterministic |
Normalize Operational Result — occurrencesau-normalize-operational-result.occurrence_count | 1UNIT | 1UNIT |
Operational Context Tokensau-normalize-operational-result.operational_context_tokens | 400TOKEN | 400TOKEN |
Prepare Retrieval Query — modeau-prepare-retrieval-query.implementation_mode | im-prepare-query-deterministic | im-prepare-query-deterministic |
Prepare Retrieval Query — occurrencesau-prepare-retrieval-query.occurrence_count | 1UNIT | 1UNIT |
Retrieval Query Tokensau-prepare-retrieval-query.retrieval_query_tokens | 180TOKEN | 180TOKEN |
Transcribe / Recognize Customer Speech — offeringDiffersau-recognize-customer-speech.commercial_offering | ctxdex:offering:agent-makers:ctxdex-realtime-2 | ctxdex:offering:agent-makers:ctxdex-speech-1 |
Customer Audio Secondsau-recognize-customer-speech.customer_audio_seconds | 150SECOND | 150SECOND |
Expected Transcript Tokensau-recognize-customer-speech.expected_transcript_tokens | 700TOKEN | 700TOKEN |
Transcribe / Recognize Customer Speech — modeDiffersau-recognize-customer-speech.implementation_mode | im-recognize-realtime-2 | im-recognize-speech-1 |
Transcribe / Recognize Customer Speech — occurrencesau-recognize-customer-speech.occurrence_count | 1UNIT | 1UNIT |
Recognition Prompt Tokensau-recognize-customer-speech.recognition_prompt_tokens | 0TOKEN | 0TOKEN |
Candidate Document Countau-rerank-candidates.candidate_document_count | 10UNIT | 10UNIT |
Candidate Document Tokensau-rerank-candidates.candidate_document_tokens | 8000TOKEN | 8000TOKEN |
Rerank Candidates — offeringau-rerank-candidates.commercial_offering | ctxdex:offering:agent-makers:ctxdex-retrieval-2-rerank | ctxdex:offering:agent-makers:ctxdex-retrieval-2-rerank |
Rerank Candidates — modeau-rerank-candidates.implementation_mode | im-rerank-model | im-rerank-model |
Rerank Candidates — occurrencesau-rerank-candidates.occurrence_count | 1UNIT | 1UNIT |
Query Countau-rerank-candidates.query_count | 1UNIT | 1UNIT |
Request Output Tokensau-rerank-candidates.request_output_tokens | 0TOKEN | 0TOKEN |
Retrieval Query Tokensau-rerank-candidates.retrieval_query_tokens | 180TOKEN | 180TOKEN |
Candidate Document Countau-search-enterprise-knowledge.candidate_document_count | 10UNIT | 10UNIT |
Candidate Document Tokensau-search-enterprise-knowledge.candidate_document_tokens | 8000TOKEN | 8000TOKEN |
Search Enterprise Knowledge — modeau-search-enterprise-knowledge.implementation_mode | im-search-local-unpriced | im-search-local-unpriced |
Search Enterprise Knowledge — occurrencesau-search-enterprise-knowledge.occurrence_count | 1UNIT | 1UNIT |
Assistant Audio Secondsau-synthesize-spoken-support-response.assistant_audio_seconds | 120SECOND | 120SECOND |
Synthesize Spoken Response — offeringDiffersau-synthesize-spoken-support-response.commercial_offering | ctxdex:offering:agent-makers:ctxdex-realtime-2 | ctxdex:offering:agent-makers:ctxdex-speech-2-realtime |
Synthesize Spoken Response — modeDiffersau-synthesize-spoken-support-response.implementation_mode | im-synthesize-realtime-2 | im-synthesize-speech-2-realtime |
Synthesize Spoken Response — occurrencesau-synthesize-spoken-support-response.occurrence_count | 1UNIT | 1UNIT |
Response Text Tokensau-synthesize-spoken-support-response.response_text_tokens | 800TOKEN | 800TOKEN |
TTS Instruction Tokensau-synthesize-spoken-support-response.tts_instruction_tokens | 200TOKEN | 200TOKEN |
Only in Native Realtime Voice
No input unique to this variantOnly in Text-Orchestrated Voice
No input unique to this variantDesign differences
What the two designs do differently, before any money is involved.
Transcribe / Recognize Customer Speech
Implementation mode
au-recognize-customer-speechSynthesize Spoken Response
Implementation mode
au-synthesize-spoken-support-responseOperations
Aligned on canonical identity, never on a display name.
In both6
- Deliver Spoken Response
op-deliver-spoken-response - Execute Required Action
op-execute-voice-support-action - Generate Support Response
op-generate-voice-support-response - Interpret & Plan Support
op-interpret-plan-voice-support - Receive & Understand Voice
op-receive-understand-voice - Retrieve Supporting Knowledge
op-retrieve-voice-support-knowledge
Only in Native Realtime Voice0
None
Only in Text-Orchestrated Voice0
None
Activities
Aligned on canonical identity, never on a display name.
In both11
- Assemble Grounded Context
au-assemble-grounded-context - Embed Query
au-embed-query - Execute Operational Tool
au-execute-operational-tool - Generate Grounded Response
au-generate-grounded-response - Interpret and Plan Request
au-interpret-plan - Normalize Operational Result
au-normalize-operational-result - Prepare Retrieval Query
au-prepare-retrieval-query - Transcribe / Recognize Customer Speech
au-recognize-customer-speech - Rerank Candidates
au-rerank-candidates - Search Enterprise Knowledge
au-search-enterprise-knowledge - Synthesize Spoken Response
au-synthesize-spoken-support-response
Only in Native Realtime Voice0
None
Only in Text-Orchestrated Voice0
None