Can I run AREX-2 27B locally?

At the default Q4_K_M, 0 of 150 phones have enough estimated usable memory; 0 also reach 3+ tokens/s. On Mac, 46 of 54 exact configurations pass using that quant or a smaller tracked fallback.

How to run AREX-2 27B locally

AREX-2 is a dense 27B research and coding agent. Q4_K_M weights are 17.2GB before runtime, context and the separate vision projector. Its iterative research workflow needs tools and feedback; loading the weights alone does not provide that workflow.

The publisher documents a recent Transformers release with Qwen3.8 support. The tracked Bartowski GGUF is a third-party conversion; AICanRun has not tested it in a phone app or measured its speed.

Publisher inference instructions ↗
Params
27B
Family
arex
Released
2026-09
Tags
chat · coding · vision · reasoning

Downloads & memory by quant

The download is only the weights — add KV cache and runtime to get what your phone actually needs. “Memory-fit” only means it can load; “usable” also requires an estimated 3+ tokens/s. Each row uses that row's quant. ★ = recommended.

QuantDownloadMemory neededPhone reality
Q2_K10.6 GB~13.6 GB4 memory-fit · 4 usable
Q3_K_M13.2 GB~16.4 GB3 memory-fit · 0 usable
IQ4_XS15.2 GB~18.5 GB3 memory-fit · 0 usable
Q4_016.1 GB~19.4 GB3 memory-fit · 0 usable
Q4_K_S16.1 GB~19.4 GB3 memory-fit · 0 usable
Q4_K_M ★17.2 GB~20.6 GB0 memory-fit
Q5_K_M20.7 GB~24.2 GB0 memory-fit
Q6_K23.6 GB~27.3 GB0 memory-fit
Q8_028.7 GB~32.6 GB0 memory-fit

AREX-2 27B at Q4_K_M on 150 phones

This table is fixed to the recommended Q4_K_M, on each phone's largest RAM option. Counts for smaller quants above will differ. All phones are shown, sorted by fit and estimated experience; tap one for the full report.

PhoneChip · RAMNeedsSpeedVerdict

~ = bandwidth-based estimate · ✓ = measured on real hardware

AREX-2 27B on 54 Mac configurations

Each Mac uses the recommended Q4_K_M at 4K context when it fits, or the largest smaller tracked quant that fits. Each row links to the full report with a macOS-specific setup guide.

MacUnified memoryQuantWorking budgetNeedsEstimated speedVerdict

FAQ

How much RAM do I need to run AREX-2 27B on a phone?

At Q4_K_M, AREX-2 27B needs ~20.6GB of usable memory (weights + KV cache + runtime). In practice that means no current Android phone or no current iPhone — iOS lets a single app use less of its RAM than Android does.

How fast does AREX-2 27B run on a flagship phone?

It doesn't fit on any of the 150 phones we track at Q4_K_M — it needs ~20.6GB of usable memory.

Can an iPhone run AREX-2 27B?

Not really: the best iPhone we track (iPhone 16 Pro Max) has ~5.2GB usable, but AREX-2 27B needs 20.6GB at Q4_K_M.

What is the best quantization of AREX-2 27B for mobile?

Q4_K_M (17.2GB download) is the size/quality sweet spot of the 9 quants available. Total memory needed is ~20.6GB once the KV cache and runtime are counted — the download size alone understates it.

How many phones can run AREX-2 27B?

0 of the 150 phones we track have enough estimated usable memory at Q4_K_M; 0 also reach our usable-speed threshold of 3 tokens/s, and 0 run smoothly. Memory-fit alone does not mean a good experience.

Which Macs can run AREX-2 27B?

46 of the 54 exact Mac configurations we track have enough estimated working memory. Each Mac uses the recommended Q4_K_M when it fits, or the largest smaller tracked quant that fits, and separates installed unified memory, the macOS reserve, and estimated speed.

What is AREX-2 27B good for on a phone?

It's tagged for chat, coding, vision, reasoning. At 27B parameters, it prioritizes answer quality over speed — expect slower decoding.

Similar models

AREX Turbo 4BQwen 3.5 27BQwen 3.6 27BBonsai 27B (1-bit)Ternary Bonsai 27B