Raon-1 mini preview runs at 68 tok/s (68.4 as recorded) on a 96 GB M2 Max, a community run reported by omlx.ai (MLX, NVFP4, 1k context). It needs about 18 GB of unified memory. It is KRAFTON's 28.8B mixture-of-experts model (4.3B active), released under the Apache 2.0 licence.
Want to run Raon-1 mini preview faster — or step up to bigger models? Find your Mac →This is a community run: a real run, reported by its source and linked below. The index records it as given and never adjusts it.
| Metric | Value |
|---|---|
| Tokens per second | community68 tok/s |
| Provenance | community run · omlx.ai |
| Quantization | NVFP4 |
| Engine | MLX |
| Time to first token | 1.062s |
| Prompt processing (prefill) | community947 tok/s at 4K context |
| Context | 1k |
| Figure dated | 2026-11-14 |
| Minimum RAM | 18 GB |
| Parameters | 28.8B MoE (4.3B active) |
| Chip bandwidth | 400 GB/s |
Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (18 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.
| Speed | Quant | Engine | Mac RAM | Context | Source | Note |
|---|---|---|---|---|---|---|
| community68.4 tok/s | NVFP4 | MLX | 96 GB | 1k | omlx.ai | 38-core GPU. oMLX v0.8.0rc1 with float16 activations, MTP/DFlash/SpecPrefill off; KRAFTON Raon-1-mini-preview-mlx (NVFP4 experts, MXFP8 output head; the page lists its quant as unknown); 18.5 GB peak. 62.2 tok/s at 4k. An earlier session on 9 Nov (v0.7.4.dev1, tr6q86gs) read 66.8. |
Measured figures only, each from its source’s own session; the index does not estimate long-context or prefill speed, so a missing cell is —. Each @ column is plain decode with that many tokens already in the prompt, from the same session as the row’s short-context figure: decode slows as the KV cache it reads grows. Prefill is prompt processing, the tokens per second a runtime reads a prompt at; with long agent contexts it sets how long you wait before the first token.
| Run | Decode (row) | @4K | Prefill | TTFT |
|---|---|---|---|---|
| MLX · NVFP4 · 96 GB Mac · omlx.ai | community68.4 tok/s | community62.2 tok/s | community947 tok/s | community1.062 s |
Ollama — the index has not verified an Ollama tag for Raon-1 mini preview, so it doesn’t print one (a guessed tag fails with “pull model manifest: file does not exist”). Search the Ollama library →
MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:
Downloads: Hugging Face repo → · Hugging Face model card →
The 68 tok/s figure above came from MLX at NVFP4; other runtimes and quants run at different speeds.
Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.
KRAFTON's Apache 2.0 preview MoE trained from scratch (21 Oct 2026): 28.8B total, 4.3B active, 262K context, text only, Korean-first across seven languages, thinking mode by default. KRAFTON's card (preview, vendor-run, GPT judge) reports GPQA Diamond 84.3, LiveCodeBench v6 81.6 and SWE-bench Verified 70.8 on mini-SWE-agent, with MMLU-Pro 76.5 below its Qwen 3.6 and Gemma 4 comparators. Official GGUFs start at 16.1 GB (IQ4_XS), so it targets 24 GB Macs. It runs in oMLX 0.7.3+ and mlx-vlm 0.7.12+ without remote code (upstream llama.cpp support is KRAFTON's open PR #32176), and community oMLX runs measured 114–129 tok/s on an M4 Max (MTP off); specs may change before the final release.
Apache 2.0 licence · 28.8B MoE · 262K context · LLMCheck capability 34/50 · see its leaderboard rank →
Estimated, not measured: the index’s minimum for Raon-1 mini preview (18 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. MLA on all 25 layers (kv_lora_rank 512 + rope 64); runtimes that expand MLA to per-head K/V use far more. A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M2 Max Mac on macOS 15, the lower default; macOS 26 allows more): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).
| Context | KV cache (16-bit) | RAM needed | RAM, q8 KV | Smallest tier that fits | Within the default GPU limit |
|---|---|---|---|---|---|
| 8K | ~0.2 GB | ~18.1 GB | ~18.1 GB | 24 GB Mac raise GPU limit | 32 GB Mac |
| 32K | ~0.9 GB | ~18.8 GB | ~18.4 GB | 24 GB Mac raise GPU limit | 32 GB Mac |
| 64K (coding-agent context) | ~1.9 GB | ~19.8 GB | ~18.9 GB | 24 GB Mac raise GPU limit | 32 GB Mac |
| 128K | ~3.8 GB | ~21.7 GB | ~19.8 GB | 32 GB Mac raise GPU limit (macOS 15) | 36 GB Mac |
Raon-1 mini preview needs about 18 GB of unified memory. These Apple Silicon Macs have enough to run it. Apple builds the higher-memory configurations to order; Amazon does not stock them.
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.
See how Raon-1 mini preview stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard