Working paper / 2026
Conversational AI LLM Bench
Abstract
We define conversational AI LLMs as the language models at the center of cascaded voice pipelines, between speech recognition and speech synthesis, that return a first token within one second: a median time to first token of 1,000 ms or less. The Conversational AI LLM Bench (CALLM) benchmarks this category. Each model operates every agent of a multi-agent front desk in healthcare, legal, and customer support, with production-length prompts, executable tools, and per-call database state. A simulated caller pursues a locked task over a simulated phone line, so the model receives only what a speech recognizer would have transcribed, errors included. A deterministic verifier scores tool calls, handoffs, and final database state; pass@1 and pass5 measure single-attempt capability and five-attempt reliability.
Clone the repository, point a model at an industry pack, and reproduce the evaluation environment.
Customer Support / T2-H3
What this task measures
The caller is on the line with Kestrel Electronics for one locked request: Marketplace headphone return. They keep repeating that ask through a callback offer, a different ticket, or an unrequested fix. We included it as a hard returns and refunds case because window and restocking math has to be spoken before a return is opened; state-law exclusions are the usual miss. The agent succeeds only if it stays on that request, reaches the outcome by following the verification then returns path, and leaves the record exactly as this case requires: the right row written or cleared, and nothing extra.
One model at a time, with a representative pass and fail conversation when both exist. GPT-4.1 is the default view.
Conversation cost $0.099
0:00 / 0:00 · scrub · trace sample · click a span to play, drag to scrub, scroll to pan
Audio
Speech Audio
Agent Audio
Handoffs
Tool Calls
Errors
Transcript
How a conversation passes
Combined pass is the logical and of three deterministic gates. Tools, handoff path, and final database state must all pass on the same conversation. One miss fails the call.
Combined
Fail
This sample fails because tools and handoff path did not match. Combined cannot pass unless every gate does.
Expected versus actual tool calls on this conversation.
Expected
Actual
Parameters
None recorded.
None recorded.
Output
+ role: verification
+ success: true
Parameters
- full_name: Tomas Ferreira
+ order_number: KE-4479002
- phone: 541-555-0146
Output
+ caller_safe_message: empty
+ data: {"record_found":true,"next_step":"Ask for the ZIP code on the order and the last four digits of the card it was paid with."}+ error_code: empty
+ ok: true
Parameters
card_last4: 2208
card_last4: 2208
postal_code: 94110
postal_code: 94110
Output
+ caller_safe_message: Look the caller up with identify_customer first.
+ data: empty
+ error_code: NO_RECORD_SELECTED
+ ok: false
Parameters
+ card_last4: 2208
+ postal_code: 94110
Output
+ caller_safe_message: empty
+ data: {"verified":true,"name":"Tomas Ferreira","tier":"plus"}+ error_code: empty
+ ok: true
Parameters
None recorded.
None recorded.
Output
+ caller_safe_message: empty
+ data: {"name":"Tomas Ferreira","tier":"plus","card_last4":"2208","membership_start":"2026-05-10","orders":[{"order_number":"KE-4479002","order_date":"2026-07-24","status":"delivered","fulfillment":"marketplace","delivered_date":"2026-07-27","delivery_date":null}],"open_service_appointments":[]}+ error_code: empty
+ ok: true
Parameters
None recorded.
None recorded.
Output
None recorded.
None recorded.
Parameters
- order_number: KE-4479002
Output
None recorded.
None recorded.
Parameters
reason_code: marketplace_seller
reason_code: marketplace_seller
Output
+ caller_safe_message: empty
+ data: {"escalation_id":1,"transferred":true,"reason_code":"marketplace_seller"}+ error_code: empty
+ ok: true
Open a row for parameters and outputs. Required tools must appear. Extra industry tools are allowed. Missing: transfer_to_returns, check_return_eligibility. Extra: verify_identity.
Full industry architecture, same graph as the industry page. The left diagram marks the expected specialist path in green. The right diagram marks the path this conversation actually took in green through the shared prefix, then red from the first wrong hop.
Expected path
transfer_to_verification → transfer_to_returns
Actual path
transfer_to_verification
The actual hangup matches the expected hangup.
The locked caller script. Copy it as written.
Open with exactly: "I want to return the Corva Studio Headphones I ordered last week." Give identity when asked. Order KE-4479002. If they say a Marketplace seller owns the policy, ask them to name the seller and get you to a person who can deal with that seller. Do not accept a Kestrel refund promise. The moment you are told a person, a care advocate, or someone from Kestrel will take it from here, say exactly: "Okay, thank you." and end the call yourself. Do not wait on the line. Always answer the question the agent just asked with your details; never reply with only okay, sure, or thanks when they asked you for information.