Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Apple / Apple Foundation Models 3

AFM 3 Core Advanced

Third-generation Apple product model co-developed with Google, with hardware and access depending on the specific deployment.

Language models (LLMs)Speech & voiceTranscription
availableClosed weightsGeneral purposeText-to-speechSpeech-to-text
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
10 of 21 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingSee billing notesApple Intelligence on-device
Output billingSee billing notesAccess and billing notes below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected
Maximum input tokens
Not collected
Maximum output
Not collected
Input modalities
Text, AudioSource for Input modalities (opens in a new tab)

Dictation and speech generation are explicit; do not infer a complete image/video input contract from the general multimodal label.

Output modalities
Text, AudioSource for Output modalities (opens in a new tab)

Speech uses a downstream detokenizer; not raw waveform output from the language decoder.

Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
SupportedSource for Streaming (opens in a new tab)

Production expressive-voice path streams synthesis; no generic public streaming multimodal endpoint asserted.

Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
2026-06-08Source for Release date (opens in a new tab)

Model announcement; product rollout followed on 14 September 2026.

Parameters
20B total; 1B–4B active depending on requestSource for Parameters (opens in a new tab)
Architecture
Instruction-Following Pruning; prompt-level expert selection, shared experts and periodic reselection; inactive weights in NANDSource for Architecture (opens in a new tab)
Languages
Not collected
Context caching
Not collected
Development partners
Apple, GoogleSource for Development partners (opens in a new tab)
Hardware requirements
Optimized for the most capable Apple silicon systems; exact per-device mapping not established hereSource for Hardware requirements (opens in a new tab)
Deployment
On-device Apple siliconSource for Deployment (opens in a new tab)
Published price
Product-integrated device inference; no standalone tariff verifiedSource for Published price (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Apple Intelligence on-device

Provider source (opens in a new tab)
API model ID
Not collected

Product or SDK model identity, not a verified public HTTP model selector.

Access
Product-integrated model. The exact request selector, checkpoint pin and independent developer access for this specific model are not established.
Regions
Supported Apple devices, OS versions, languages and regions only. A device-local model is not a guarantee of access in every market.

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

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.

Apple Intelligence on-device

Not verified
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

What the measurements say

Expressive voices: general MOS

Apple · vendor-reported
4.15/ 5

Apple human evaluation at 1B active parameters; previous production TTS scored 3.87. Feature-level audio pipeline result, not an independent benchmark.

Evaluation source (opens in a new tab)

Expressive voices: conversational MOS

Apple · vendor-reported
4.24/ 5

Apple human evaluation at 1B active parameters; previous production TTS scored 3.82. Not a word-error-rate metric.

Evaluation source (opens in a new tab)

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

Apple documents this proprietary product model. No matching unrestricted public checkpoint was verified in the reviewed Apple weight inventory. Closed-weight classification is inferred for this product identity; OS delivery and member-gated model assets do not imply an open-weight release.

Classification evidence (opens in a new tab)

Lifecycle dates

Model family announced8 Jun 2026Source (opens in a new tab)
Apple Intelligence product rollout begins14 Sept 2026Source (opens in a new tab)

Sources reviewed