Zoom’s internally trained multilingual small language model, named 2.3B in its official benchmark tables. Public inference access and context limits are not documented.
No public inference selector documented for this research variant.
Access
Research-only inventory record. The publication describes future AI Studio customization and internal evaluations; no public endpoint or numeric inference rate is documented.
Regions
Not collected
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Current service terms are linked for provider context; no original research/evaluation retention agreement is inferred.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
Zoom research · no public standalone inference
Not verified
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
How it works
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
Minimum reusable input
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
Lifetime & refresh
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
How to enable it
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
Usage fields
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
Billing rules
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
Limits & data boundaries
Not collected
No model-specific discounted prompt-cache contract was verified. Audio buffering, S3 job storage and orchestration memory are not treated as prompt caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
What the measurements say
MMLU
Zoom · provider-reported
65.1% accuracy
Zoom internal evaluation reported February 19, 2025; Lighteval with 5 shots. Not an independent evaluation and not a score for the specialized adaptations.
Zoom internal evaluation reported February 19, 2025; Lighteval with 5 shots. Not an independent evaluation and not a score for the specialized adaptations.
Zoom internal evaluation reported February 19, 2025; Lighteval with 2 shots. Not an independent evaluation and not a score for the specialized adaptations.
Zoom internal evaluation reported February 19, 2025; Lighteval with 3 shots. Not an independent evaluation and not a score for the specialized adaptations.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Zoom describes training and customizing its own SLM internally; the official tables name Zoom-SLM-2.3B. Later research distinguishes its proprietary native SLMs from external open- and closed-weight models. No public weight release or independently callable research checkpoint is established.
