ZAZUMODELSBLOG

Model library · Alibaba · 9B

Qwen3.5-9B

From the transcripts Top score, but it reasons out loud at the customer: policy checklists, "wait, actually", monologues up to 46,000 words. Not usable as-is.

MiniMax-strict customer

Customer simulator: minimax/minimax-m3.1-flash (strict customer patch).

2026-10-28

openrouterfp8reasoning off
Domainpass^1Higher kSimsInfra errorsAgent cost / taskTranscripts
AirlineInstruction following74.0%—500$0.0135View →
RetailTool calling78.1%—1140$0.0095View →
TelecomReasoning99.1%—1140$0.0144View →

GPT-5.2 customer

Customer simulator: gpt-5.2 (reasoning_effort=low).

2026-10-17

self-hostedbf16reasoning off
Domainpass^1Higher kSimsInfra errorsAgent cost / taskTranscripts
AirlineInstruction following76.0%^2 68.3% · ^3 64.5% · ^4 62.0%2000—View →
RetailTool calling77.2%^2 66.5% · ^3 59.6% · ^4 55.3%4560—View →
TelecomReasoning97.4%^2 94.7% · ^3 92.1% · ^4 89.5%4560—View →

GLM-5.3 Flash customer

Customer simulator: glm-5.3-flash (gpt-oss-120b on *_fixoss reruns).

2026-10-07

self-hostedreasoning off
Domainpass^1Higher kSimsInfra errorsAgent cost / taskTranscripts
RetailTool calling84.4%—9618—View →