Models / Ornith-1.5-35B-A3B

Ornith-1.5-35B-A3B

MoE enthusiast
HuggingFace → HuggingBay → Ornith AI → Find for my hardware

~36B total, ~3B active per token (MoE) - the local-favorite Ornith-1.5 size, released 2026-08-18. Activates only 3B parameters per token yet outperforms dense models like Gemma 4-31B and Qwen 3.6-35B on agentic coding. 262K context, MIT-licensed on HuggingFace at ornith-ai/Ornith-1.5-35B-A3B (GGUF, FP8, and NVFP4 quantizations).

  • Coding (vendor self-reported): Terminal-Bench 2.1 67.8, SWE-bench Verified 79.0, SWE-bench Pro 59.6, NL2Repo 46.2.
  • Reasoning: HLE 25.6 (no tools) / 33.4 (with tools), GPQA-Diamond 89.2.
  • Agentic: MCP-Atlas 70.2, Toolathlon-Verified 48.7, ClawEval 72.5.

Local-friendly. Runs on enthusiast-class GPUs via GGUF quantizations; a strong open-weight pick for local agentic coding. Vendor benchmarks are claims pending independent replication.

coding reasoning agentic
Parameters
36.0B
Context
262k
License
mit
Developer
Ornith AI
Origin
🇺🇸 USA
Released
Aug 2026

Benchmark scores

Vendor-reported - from the developer's own model card / tech report

GPQA-Diamond
89.2
HLE
25.6
MCP-Atlas
70.2
NL2Repo-Bench
46.2
SWE-bench Pro
59.6
SWE-bench Verified
79.0
Terminal-Bench 2.1
67.8
Toolathlon Verified
48.7

Vendor-reported - from the developer's own model card / tech report

Ran this model on your own hardware? Join free and add your measured tok/s to the community numbers.

Related models

Ornith-1.5-397B
403.0B workstation 248GB min RAM Aug 18, 2026
Ornith-1.5-9B
9.0B consumer 6GB min RAM Aug 18, 2026
Ornith-1.0-35B superseded by this
35.0B 3.0B active enthusiast no local build Jun 25, 2026
Ornith-1.0-397B superseded
397.0B workstation no local build Jun 25, 2026
Ornith-1.0-9B superseded
9.0B consumer no local build Jun 25, 2026

Save your hardware and every model page answers the real question: will it run on your machine, and how fast?

Join free - save your rig →

Run it locally

Per-quant memory needs and a static "can you run it?" reference - no rig entry required

Q4_K_M
21.7GB 22.0GB min 24.0GB rec
Balanced - the usual local sweet spot
Q5_K_M
25.3GB 26.0GB min 28.0GB rec
High quality, larger than Q4
Q6_K
29.2GB 30.0GB min 32.0GB rec
Near-lossless, large
Q8_0
37.8GB 38.0GB min 42.0GB rec
Near-lossless
BF16
71.1GB 72.0GB min 80.0GB rec
Full quality, largest

The reference hardware

Schematic of the NVIDIA Jetson Orin NX 16GB reference rig - 16GB unified memory, 102 GB/s aggregate bandwidth
Schematic of the Single GTX 1080 Ti (11GB) reference rig - 11GB VRAM, 484 GB/s aggregate bandwidth
Schematic of the 4x H100 80GB (320GB) reference rig - 320GB VRAM, 13400 GB/s aggregate bandwidth
Schematic of the NVIDIA DGX Station 748GB reference rig - 748GB unified memory, 8000 GB/s aggregate bandwidth
Schematic of the 8x RTX 3090 rack (192GB) reference rig - 192GB VRAM, 7489 GB/s aggregate bandwidth
Schematic of the 4x RTX 5090 (128GB) reference rig - 128GB VRAM, 7168 GB/s aggregate bandwidth
Schematic of the AMD Instinct MI300X (192GB) reference rig - 192GB VRAM, 5324 GB/s aggregate bandwidth
Schematic of the 4x RTX 4090 (96GB) reference rig - 96GB VRAM, 4032 GB/s aggregate bandwidth
Schematic of the 2x RTX 5090 (64GB) reference rig - 64GB VRAM, 3584 GB/s aggregate bandwidth
Schematic of the 2x RTX 3090 (48GB) reference rig - 48GB VRAM, 1872 GB/s aggregate bandwidth
Schematic of the Single RTX 5090 (32GB) reference rig - 32GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the RTX PRO 6000 Blackwell (96GB) reference rig - 96GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 192GB reference rig - 192GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 512GB reference rig - 512GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Single RTX 4090 (24GB) reference rig - 24GB VRAM, 1008 GB/s aggregate bandwidth
Schematic of the MacBook Pro M5 Max 128GB reference rig - 128GB unified memory, 614 GB/s aggregate bandwidth
Schematic of the Dual EPYC 9004 + 768GB DDR5-4800 reference rig - 768GB unified memory, 460 GB/s aggregate bandwidth
Schematic of the DGX Spark 128GB unified reference rig - 128GB unified memory, 273 GB/s aggregate bandwidth
Schematic of the Ryzen AI Max+ 395 128GB reference rig - 128GB unified memory, 256 GB/s aggregate bandwidth
Schematic of the Jetson AGX Orin 64GB reference rig - 64GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-3200 + 2x RTX 3090 reference rig - 560GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-2400 + 2x RTX 3090 reference rig - 560GB unified memory, 153 GB/s aggregate bandwidth

22 reference configs, drawn in-house. Scroll for more.

Can you run it? - reference rigs

Rig Q4_K_M Q5_K_M Q6_K Q8_0 BF16
NVIDIA Jetson Orin NX 16GB no -> cloud no -> cloud no -> cloud no -> cloud no -> cloud
Single GTX 1080 Ti (11GB) no -> cloud no -> cloud no -> cloud no -> cloud no -> cloud
4x H100 80GB (320GB) fast 4070.6t/s fast 3492.0t/s fast 3026.0t/s fast 2338.1t/s fast 1243.4t/s
NVIDIA DGX Station 748GB fast 2430.2t/s fast 2084.8t/s fast 1806.6t/s fast 1395.9t/s fast 742.3t/s
8x RTX 3090 rack (192GB) fast 2275.2t/s fast 1951.8t/s fast 1691.3t/s fast 1306.8t/s fast 695.0t/s
4x RTX 5090 (128GB) fast 2177.5t/s fast 1868.0t/s fast 1618.7t/s fast 1250.7t/s fast 665.1t/s
AMD Instinct MI300X (192GB) fast 1617.6t/s fast 1387.6t/s fast 1202.5t/s fast 929.1t/s fast 494.1t/s
4x RTX 4090 (96GB) fast 1224.8t/s fast 1050.7t/s fast 910.5t/s fast 703.5t/s fast 374.1t/s
2x RTX 5090 (64GB) fast 1088.7t/s fast 934.0t/s fast 809.4t/s fast 625.3t/s offload
2x RTX 3090 (48GB) fast 568.8t/s fast 487.9t/s fast 422.8t/s fast 326.7t/s offload
Single RTX 5090 (32GB) fast 544.4t/s fast 467.0t/s fast 404.7t/s offload no -> cloud
RTX PRO 6000 Blackwell (96GB) fast 544.4t/s fast 467.0t/s fast 404.7t/s fast 312.7t/s fast 166.3t/s
Mac Studio M4 Ultra 192GB fast 361.9t/s fast 310.4t/s fast 269.0t/s fast 207.9t/s fast 110.5t/s
Mac Studio M4 Ultra 512GB fast 361.9t/s fast 310.4t/s fast 269.0t/s fast 207.9t/s fast 110.5t/s
Single RTX 4090 (24GB) fast 306.2t/s offload offload offload no -> cloud
MacBook Pro M5 Max 128GB fast 203.5t/s fast 174.6t/s fast 151.3t/s fast 116.9t/s fast 62.2t/s
Dual EPYC 9004 + 768GB DDR5-4800 fast 140.0t/s fast 120.1t/s fast 104.1t/s fast 80.4t/s fast 42.8t/s
DGX Spark 128GB unified fast 82.9t/s fast 71.1t/s fast 61.7t/s fast 47.6t/s fast 25.3t/s
Ryzen AI Max+ 395 128GB fast 77.8t/s fast 66.7t/s fast 57.8t/s fast 44.7t/s fast 23.8t/s
Jetson AGX Orin 64GB fast 62.2t/s fast 53.4t/s fast 46.3t/s fast 35.7t/s no -> cloud
Epyc + 512GB DDR4-3200 + 2x RTX 3090 fast 62.2t/s fast 53.4t/s fast 46.3t/s fast 35.7t/s ok 19.0t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 fast 46.7t/s fast 40.0t/s fast 34.7t/s fast 26.8t/s ok 14.3t/s

Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.

Download options

Q4_K_M official -5% vs fp16
21.7GB dl 22.0GB min 24.0GB rec
REC RAM vs largest quant
21.7GB Q4_K_M weights + overhead + KV cache; ~22GB to run in RAM
GGUF on HF →
Q5_K_M official -3% vs fp16
25.3GB dl 26.0GB min 28.0GB rec
REC RAM vs largest quant
GGUF on HF →
Q6_K official -1% vs fp16
29.2GB dl 30.0GB min 32.0GB rec
REC RAM vs largest quant
GGUF on HF →
Q8_0 official -1% vs fp16
37.8GB dl 38.0GB min 42.0GB rec
REC RAM vs largest quant
GGUF on HF →
BF16 official
71.1GB dl 72.0GB min 80.0GB rec
REC RAM vs largest quant
GGUF on HF →

Or run it in the cloud

No per-token API provider pricing tracked for Ornith-1.5-35B-A3B yet. For flagship list prices, see the calculator.

PRICE HISTORY

Inference cost over time

Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.

Loading price history...