coreai-model-zoo
Open-source repo of pre-converted .aimodel bundles and recipes for Apple Core AI on iOS 27 and macOS 27.
If you are shipping Core AI into a Swift app, start here. The recipe-plus-verification loop is the part DIY conversion never gives you — zoo_convert.py reruns the conversion, zoo_verify.py checks the bundle against its source, and one-line CoreAIKit 0.4.1 loading gets a chat model answering fast. The recorded 94 tok/s for Qwen3-8B 4-bit on an M4 Max GPU, against MLX's 90 under the same protocol, is a real signal rather than a marketing number. Just read the model card before you commit: a Mac run is not evidence for the iPhone variant, and parity proof strength varies by model. If you need Android, Windows or cloud deployment, use MLX or a hosted API instead.
Verified 7d ago · liveness 78/100 · cite: rightaichoice.com/tools/coreai-model-zoo
- iOS and macOS developers shipping on-device AI with Apple's Core AI runtime
- Swift teams that want a chat model answering in an app without hand-rolling a conversion pipeline
- Engineers who need to audit or rebuild a model bundle before it ships
- AI engineers converting PyTorch LLMs to .aimodel who want a working reference
- Teams deploying the same model to Android, Windows or cloud servers
- Anyone needing commercial support, SLAs or a vendor escalation path
- Developers not yet on the iOS 27 / macOS 27 beta toolchain
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip CoreAI-Model-Zoo if you need the same model running on Android, Windows or cloud servers, or you require a vendor SLA — this repo targets Apple's Core AI on iOS 27 and macOS 27 only.
Your first CoreAIKit download is about 352 MB on Mac and roughly 456 MB for the separate iPhone variant, so cold-start installs and CI caches carry real bandwidth and storage cost.
The repo itself is $0 with no tiers, which undercuts any paid on-device inference service for Apple platforms. The real cost comparison is engineering time: versus paying for a hosted API, you spend Swift integration hours; versus MLX, which is also free, the difference is that this repo hands you gated bundles and app-shaped loading paths instead of a broader research surface you assemble yourself.
In short
coreai-model-zoo — Open-source repo of pre-converted .aimodel bundles and recipes for Apple Core AI on iOS 27 and macOS 27. Best for iOS and macOS developers shipping on-device AI with Apple's Core AI runtime, Swift teams that want a chat model answering in an app without hand-rolling a conversion pipeline, Engineers who need to audit or rebuild a model bundle before it ships. Free to use.
What's new in coreai-model-zoo
Checked 7 days agoAcross the latest 2 updates: 1 feature update and 1 changelog entry.
Benchmark recorded: Qwen3-8B 4-bit at 94 tok/s on M4 Max GPU vs MLX 90
Under apple-silicon-llm-bench on macOS 27 beta 26A5353q, Qwen3-8B 4-bit decodes at 94 tok/s on an M4 Max GPU against MLX at 90 on the same protocol.
CoreAIKit 0.4.1 with built-in catalog, ChatDemo app and chat-cli
The published 0.4.1 release adds a built-in model catalog so you resolve an entry like qwen3-0.6b and get a ChatSession, plus the ChatDemo GUI app and chat-cli runner.
What people actually say about coreai-model-zoo — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
22 mentions across 2 sources (Hacker News, YouTube) · researched Aug 15, 2026.
Average across the 2 sources that answered — each source counts once, not each post.
- +Free and open-source, with 57 pre-converted .aimodel files.
- +Covers a wide range of model types: LLMs, VLMs, OCR, ASR, TTS, and generation.
- +Reproducible conversions with recipes and zoo_convert.py for auditability.
- +One-line Swift loading via CoreAIKit is a real time-saver.
- +Custom Metal kernels for GPU/ANE acceleration are included.
- −No Windows or NVIDIA support; strictly Apple ecosystem only.
- −Performance on expensive Macs is questioned by some users.
- −No official support channels or documentation for common issues.
- −Requires advanced Swift and Core AI knowledge to use effectively.
- −Community feedback is sparse; few real-world validation reports.
- • Time to learn Core AI and integrate the zoo into existing projects
- • Potential cloud or infrastructure costs if you deploy at scale (but the zoo itself is local)
Viability Score
How well maintained and how widely used is coreai-model-zoo? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- 62 pre-converted .aimodel bundles for Apple Core AI on iOS 27 and macOS 27
- One-line Swift loading through CoreAIKit 0.4.1's built-in model catalog
- Per-model conversion recipes in models/<model>/recipe.toml, rerunnable with zoo_convert.py run <name>
- zoo_verify.py checks a published bundle against its source model
- Every bundle gated against its source model before release
- Model cards state what was measured on which hardware and the strength of parity proof
- KitLanguageModel integration for Apple FoundationModels LanguageModelSession
- Chat ports including qwen3-0.6b and Qwen3-8B 4-bit
- Vision-language ports including Qwen3-VL and MiniCPM-V
- ASR with Whisper and Qwen3-ASR
- Text-to-speech with Kokoro-82M and VoxCPM 0.5B
- Generative tasks: image, video and music generation
- OCR and forecasting model ports
- Custom Metal kernels for GPU and Neural Engine acceleration
- iPhone-tier models device-measured; large models labeled Mac-only
About coreai-model-zoo
CoreAI-Model-Zoo is a free open-source GitHub repository that skips the conversion tax for Apple on-device AI. It ships downloadable .aimodel bundles for Core AI — Apple's on-device ML runtime in iOS 27 and macOS 27, and the successor to Core ML — covering chat LLMs (including Qwen3-0.6B and Qwen3-8B 4-bit), vision-language ports like Qwen3-VL and MiniCPM-V, OCR, speech with Whisper/Qwen3-ASR and TTS with Kokoro-82M and VoxCPM, plus image, video and music generation. Every bundle carries the recipe that produced it in models/<model>/recipe.toml, so you can rerun it with zoo_convert.py run <name> instead of trusting a blob. Verification is a separate step: zoo_verify.py checks a published bundle against its source model, and each card states what was measured on which hardware — iPhone tiers are device-measured, large models are labeled Mac-only and say so. The repo reports Qwen3-8B 4-bit decoding at 94 tok/s on an M4 Max GPU versus MLX at 90 under the same protocol (apple-silicon-llm-bench, macOS 27 beta 26A5353q, 2026-06-11). Getting a model into an app runs through CoreAIKit 0.4.1, a Swift package with a built-in catalog: pick qwen3-0.6b, get a ChatSession, call respond(to:). The first download is roughly 352 MB on Mac, about 456 MB for the separate iPhone variant. The release also includes the ChatDemo app and a chat-cli runner. It is built for Apple developers, and for PyTorch and MLX users who just want a working model on an iPhone or Mac.
Behind the Verdict
The value here is discipline, not novelty. Anyone can export a PyTorch model through Apple's coreai-torch — LLMs use coreai.llm.export — and get a .aimodel file. What almost nobody does is publish the recipe alongside the artifact and gate each release. CoreAI-Model-Zoo does both: models/<model>/recipe.toml reproduces the bundle, zoo_convert.py lists, shows, dry-runs and reruns conversions, and where the shipped configuration could not be recovered from the repository, the recipe says so rather than guessing. That honesty is the reason to trust the catalog. The catalog covers more ground than the chat-model framing suggests. There are chat LLMs, vision-language ports (Qwen3-VL, MiniCPM-V), OCR and forecasting ports, ASR through Whisper and Qwen3-ASR, TTS through Kokoro-82M and VoxCPM 0.5B, and generative image, video and music tasks that use their own Kit APIs rather than the chat path. Custom Metal kernels are included for GPU and Neural Engine acceleration. The weak spots are real. Large models are Mac-only and the cards say so — do not expect an iPhone port of everything. Community ports arrive from the contributor's own Hugging Face namespace, credited by name, so per-model provenance varies and you should read the validation record for the specific model you pick. Tool calling through KitLanguageModel depends on the model's dialect, and guided generation needs a compatible sequential engine, so FoundationModels integration is not uniformly plug-and-play. Conversion and verification depend on the repo's own tooling rather than a hosted service, which means your CI has to run them. Where it fits: Swift and iOS/macOS teams who want an on-device chat or speech model in an app this week and who are willing to read a model card. Where it does not: cross-platform teams, anyone wanting a supported SLA, and teams still off the iOS 27 / macOS 27 beta toolchain. Compared with MLX, the trade is narrowness for depth — MLX gives a broader research surface, this gives a gated, reproducible catalog plus Swift loading paths that assume an app.
Researching coreai-model-zoo? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas coreai-model-zoo actually fits — and what changes day-one when you adopt it.
Add CoreAIKit 0.4.1 via Swift Package Manager, resolve qwen3-0.6b from the built-in catalog, create a ChatSession and call respond(to:).
Outcome: The Kit downloads the matching iPhone variant once and caches it; prompts and inference stay on the device with no server round trip.
Read models/<model>/recipe.toml, reproduce the export with coreai.llm.export, then run zoo_convert.py on the result and check the output against the source model with zoo_verify.py.
Outcome: You end up with a .aimodel bundle whose provenance you can defend, plus the recipe stored next to it for the next rebuild.
Open the model card for the specific model, check which hardware the numbers were measured on and how strong the parity proof is, then rerun zoo_verify.py yourself.
Outcome: You either clear the bundle for production or catch that the only measurements available are Mac-only before it reaches a device.
Use Cases
- Run qwen3-0.6b locally in a Swift chat app using the built-in CoreAIKit catalog entry
- Decode Qwen3-8B 4-bit on an M4 Max GPU through Core AI at the recorded 94 tok/s
- Ship a vision-language feature with the Qwen3-VL or MiniCPM-V port
- Add on-device speech recognition with Whisper or Qwen3-ASR
- Add on-device text-to-speech with Kokoro-82M or VoxCPM 0.5B
- Rebuild a bundle from its recipe.toml to audit what you are shipping
- Re-run zoo_verify.py to check a published bundle against its source model
- Study conversion gotchas to port your own PyTorch or MLX model to Core AI
Models Under the Hood
as of 2026-09-25
Limitations
- The zoo targets Apple Core AI on iOS 27 and macOS 27, and large models are labeled Mac-only rather than iPhone-runnable — a Mac benchmark is not evidence for the iPhone variant.
- Community ports are served from contributors' own Hugging Face namespaces, so provenance varies and you must read the per-model validation record.
- Tool calling through KitLanguageModel depends on the model's dialect, and guided generation needs a compatible sequential engine.
- Conversion and verification run through the repo's own tooling (zoo_convert.py, zoo_verify.py) rather than a hosted service, so you carry that in your own CI.
- The ChatDemo build pins a specific beta DEVELOPER_DIR (Xcode 27.0.0 Beta 5) unless you adjust it.
as of 2026-10-01
Verification history
We have re-verified coreai-model-zoo 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published coreai-model-zoo tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free (open-source repo)
$0
Ideal for
iOS and macOS developers, indie Swift teams and AI engineers who need pre-converted Core AI bundles and are willing to read model cards and run their own verification.
What this tier adds
Starting tier and only tier — $0, includes all 62 .aimodel bundles, the recipes in models/<model>/recipe.toml, zoo_convert.py and zoo_verify.py, CoreAIKit 0.4.1, the ChatDemo app and chat-cli, and the knowledge base.
Where the pricing makes sense
The company stage and team size where coreai-model-zoo's pricing actually pencils out — and where peers do it cheaper.
The repo itself is $0 with no tiers, which undercuts any paid on-device inference service for Apple platforms. The real cost comparison is engineering time: versus paying for a hosted API, you spend Swift integration hours; versus MLX, which is also free, the difference is that this repo hands you gated bundles and app-shaped loading paths instead of a broader research surface you assemble yourself.
Setup time & first value
How long it actually takes to get something useful out of coreai-model-zoo — broken out by persona, not the marketing-page minute.
Swift integration: add CoreAIKit 0.4.1 and get a first answer within an hour, plus one ~352 MB Mac download (about 456 MB for the iPhone variant). Conversion work: budget hours to days per model depending on whether recipe.toml recovers the shipped configuration. Verification: minutes to rerun zoo_verify.py per bundle.
Switching to or from coreai-model-zoo
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From hand-rolled Core ML conversion scripts: swap your pipeline for models/<model>/recipe.toml plus zoo_convert.py run <name> and keep zoo_verify.py as the release gate.
- →From MLX prototypes: port the model through coreai.llm.export into a .aimodel bundle and load it via the CoreAIKit built-in catalog instead of an MLX runtime.
- →From CoreML-Models: this repo is the stated successor, so carry over your model choices and adopt the recipe-plus-gate workflow.
- ↗To MLX: export the same weights into an MLX runtime when you want a broader research surface than a gated app-shaped catalog.
- ↗To a hosted inference API: move the chat path server-side if you need Android, Windows or cloud coverage the Core AI bundles do not provide.
- ↗To Apple's own coreai-torch directly: keep the export tooling and drop the catalog when you only need one model and no gate records.
Integrations
Resources & Guides
- Resourcegithub.com
Coreai Model Zoo · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
README · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
PORTING · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
BENCHMARKS · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
CONTRIBUTING · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
SECURITY · coreai-model-zoo
Helpful link from github.com
- Resourcegithub.com
Coreai Kit · coreai-model-zoo
Helpful link from github.com
Tutorials & Learning
YouTube returned 6 videos for “coreai-model-zoo”, and we withheld 5: 5 did not mention coreai-model-zoo. Showing the 1 we can prove is about coreai-model-zoo.
Official links
Tools that pair well with coreai-model-zoo
Common stack mates teams adopt alongside coreai-model-zoo, with the specific reason each pairing earns its keep.
Cortex.cpp
Free, open-source desktop app to run 123 HuggingFace models locally or route prompts to Claude, GPT, Gemini and DeepSeek with your own API keys
RWKV Runner
Free, open-source desktop app for running and fine-tuning RWKV RNN language models locally with infinite context.
Vmlx
Free, MIT-licensed local LLM inference for Apple Silicon Macs, with SSD-backed prefix caching and native OpenAI and Anthropic APIs.
Featured Head-to-Head Comparisons
Coreai Model Zoo vs Spider Cloud
Spider Cloud and CoreAI-Model-Zoo serve completely different needs: Spider Cloud is a cloud API for web data extraction (RAG, AI agents), while CoreAI-Model-Zoo is a free, on-device model repository for Apple developers. If you need live web data for AI pipelines at low cost, go with Spider Cloud. If you're building local AI apps on iPhone/Mac with optimized Core AI models, CoreAI-Model-Zoo is the obvious choice.
Coreai Model Zoo vs Voyage Ai
These tools serve completely different needs. Choose Voyage AI if you run an enterprise RAG pipeline needing domain-tuned embeddings and rerankers, especially for finance/legal; its 32K context and low-dim vectors reduce storage cost. Choose coreai-model-zoo if you're an Apple developer deploying LLMs on-device on iOS 27 / macOS 27 — it's free, open-source, and includes pre-converted models with GPU/ANE acceleration. No overlap in target audience.
Coreai Model Zoo vs Temporal Ai
These tools solve entirely different problems. Choose Temporal AI if you need a robust orchestration platform to build reliable, fault-tolerant AI agents and long-running workflows — it's overkill for simple tasks but essential for mission-critical systems. Choose CoreAI-Model-Zoo if you are an iOS/macOS developer targeting on-device AI with Apple's latest hardware and need ready-to-use, optimized models. There is no meaningful overlap; your choice depends on whether your bottleneck is workflow reliability or on-device model deployment.
Alternatives to coreai-model-zoo
View all
Cortex.cpp
Free, open-source desktop app to run 123 HuggingFace models locally or route prompts to Claude, GPT, Gemini and DeepSeek with your own API keys
RWKV Runner
Free, open-source desktop app for running and fine-tuning RWKV RNN language models locally with infinite context.
Frequently Asked Questions
Used coreai-model-zoo? Help shape our editorial sentiment research.