The first open model in KRAFTON AI's Raon family, trained from scratch (preview release 2026-10-22, Apache-2.0, official size designation 30B-A4B): 28.8B total parameters, 4.3B active (6 of 128 routed experts plus 1 shared). Hybrid attention is the headline — 18 of 25 layers use Raon Delta Attention (linear attention that decouples the erase and write gates of Kimi Delta Attention, with a fixed-size state) and 7 use Gated MLA (KV cache of only ~8KiB per token in BF16) in a 3:1 layout; 262,144-token context. Multilingual Korean/English/Japanese/Chinese/Spanish/French/Thai with native tool calling (Hermes Agent, Pi and OpenCode already integrated). Official quantizations are MXFP8 and NVFP4, and **NVFP4 runs on a single NVIDIA RTX 5090 (32GB)**; a DFlash2 draft model ships alongside (up to 2.18x higher output throughput measured on a B300). Weight repos are request-to-access with automatic approval; official GGUFs: Q4_K_M 19.7GB, Q6_K 25.8GB, Q8_0 30.7GB. Note: preview release — specs and results may change before the final version.
0
Quantized builds
24
Benchmark records
1
Discussions
Quantized builds
Community and official quantized releases; "linked records" counts only benchmarks tied to that exact release.
No quantized builds cataloged yet.
Benchmark data
23 configurations
| Hardware | Framework | Quant | Decode | VRAM used | Samples | Evidence |
|---|---|---|---|---|---|---|
| NVIDIA RTX 5090 | Ollama | Q2_K | 351.2 tok/s | 30.4 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | llama.cpp | IQ2_M | 46.3 tok/s | 33.55 GB | 1 | L2 Cross-framework verified |
| AMD Ryzen AI Max+ 395 | Ollama | IQ2_M | 44.2 tok/s | 33.55 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | llama.cpp | IQ2_XS | 104 tok/s | 31.13 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | llama.cpp | IQ4_XS | 87 tok/s | 46.85 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | Ollama | IQ1_S | 109.9 tok/s | 27.5 GB | 1 | L2 Cross-framework verified |
| Apple M5 Max | Ollama | Q5_K_S | 61.5 tok/s | 56.53 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | llama.cpp | Q8_0 | 76.1 tok/s | 71.65 GB | 1 | L2 Cross-framework verified |
| Apple M5 Ultra | Ollama | Q4_K_S | 118.2 tok/s | 49.28 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | llama.cpp | Q6_K | 343.5 tok/s | 62.58 GB | 1 | L2 Cross-framework verified |
| NVIDIA H100 80GB HBM3 | Ollama | Q3_K_S | 313.6 tok/s | 41.41 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | Ollama | IQ4_XS | 54.6 tok/s | 46.85 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | Ollama | Q3_K_M | 58.6 tok/s | 43.23 GB | 1 | L2 Cross-framework verified |
| NVIDIA H20 | Strata | Q5_K_M | 68.6 tok/s | 58.35 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | vLLM | IQ3_S | 304.1 tok/s | 42.02 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | vLLM | IQ4_XS | 310.1 tok/s | 46.85 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX PRO 6000 Blackwell 96GB | Ollama | Q4_K_M | 227.9 tok/s | 51.69 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | llama.cpp | IQ2_M | 155.9 tok/s | 33.55 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | llama.cpp | IQ3_S | 130 tok/s | 42.02 GB | 1 | L2 Cross-framework verified |
| Apple M2 Ultra | Ollama | Q4_K_M | 110.5 tok/s | 51.69 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | llama.cpp | Q4_K_M | 47.8 tok/s | 51.69 GB | 2 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | IQ4_XS | 49.6 tok/s | 46.85 GB | 1 | L2 Cross-framework verified |
| NVIDIA RTX Spark | Ollama | Q4_K_M | 45 tok/s | 51.69 GB | 1 | L2 Cross-framework verified |
24 of these records are not linked to a specific quantized build; the full set is on the search page.