This model is a fine-tuned version of DuyTa/Cyber-F1. It has been trained using TRL. This model was trained with SFT. - PEFT 0.21.0
Open-weight model · Text generation
Longleaf-35B-A3B
by Red Hills Media redhillsmediafl/Longleaf-35B-A3B
Longleaf-35B-A3B is an open-weight model for text generation from Red Hills Media, released under Apache License 2.0. It has 35.1B parameters and a 262,144-token context. At 16-bit it needs about 84.3 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
Longleaf is our local model for coding and agent work: repository edits, debugging, structured tools, and clear answers. Built on Qwen3.6-35B-A3B. - 35B total / 3B active parameters. - MLX 4-bit weights. - 262,144-token configured context.
Parameters35.1B
Context262,144
Weights20.4 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads—
Runs On
What it takes to serve Longleaf-35B-A3B (35.1B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 70.2 GB | 84.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x MI325X $2.00 · 1x MI355X $2.59 |
| 8-bit | 35.1 GB | 42.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 17.6 GB | 21.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 19, 2026.
Longleaf-35B-A3B on every accelerator the SAVRN Index prices, at every precision
Model Card
By Red Hills Media, published under apache-2.0, revision dd4b62ee9c24.
Longleaf is our local model for coding and agent work: repository edits, debugging, structured tools, and clear answers. Built on Qwen3.6-35B-A3B. - 35B total / 3B active parameters. - MLX 4-bit weights. - 262,144-token configured context. EvalPlus 0.3.1, 164 tasks, pass@1. Thinking off, greedy sampling, 2,048-token output cap, 262,144-token configured context. Generation used our M1 Ultra 231 / 198 / 195 tokens, respectively. CLI sample: five Terminal-Bench 2 tasks on ARM64, thinking off. Per task: 300 seconds, 24 model calls, 32 tool calls, 2,048 output tokens per call. Our M1 Ultra Splash runtime, medium thinking, model already loaded: Read, edit, and test completed in about 6 seconds…
Read Red Hills Media's full model card
Longleaf is our local model for coding and agent work: repository edits, debugging, structured tools, and clear answers.
Built on Qwen3.6-35B-A3B.
- 35B total / 3B active parameters.
- MLX 4-bit weights.
- 262,144-token configured context.
- Development hardware: M1 Ultra, 64 GB unified memory, 48-core GPU.
- Splash thinking levels: none, low, medium, high, xhigh, max.
Coding Benchmarks
EvalPlus 0.3.1, 164 tasks, pass@1. Thinking off, greedy sampling, 2,048-token output cap, 262,144-token configured context. Generation used our M1 Ultra Splash runtime.
| Model | HumanEval | HumanEval+ | Median generation time | CLI sample (5 tasks) |
|---|---|---|---|---|
| Qwen3.6 base | 95.1% | 90.9% | 1.19 s | 1/5 |
| Longleaf | 96.3% | 91.5% | 0.97 s | 1/5 |
| Longleaf Abliterated | 95.7% | 90.9% | 0.94 s | 1/5 |
Prompts: 159 tokens median, 77–441 range. Median output lengths: 231 / 198 / 195 tokens, respectively.
CLI sample: five Terminal-Bench 2 tasks on ARM64, thinking off. Per task: 300 seconds, 24 model calls, 32 tool calls, 2,048 output tokens per call.
Pi Coding Run
Our M1 Ultra Splash runtime, medium thinking, model already loaded:
| Model calls | Prompt tokens per call | Output tokens | Decode | First token |
|---|---|---|---|---|
| 4 | 1,731–2,236 | 439 total | 104–130 tokens/s | 1.3 s first; 0.2–0.3 s cached |
Read, edit, and test completed in about 6 seconds, with 4/4 tests passing.
Measured on a cached coding task.
Run
pip install -U mlx-lm
mlx_lm.generate --model redhillsmediafl/Longleaf-35B-A3B --prompt "Help me debug this code."
License
Apache 2.0. This release contains modified model weights, based on Qwen3.6-35B-A3B and the MLX conversion. Review generated code before running it and restrict agent tool permissions.
Configuration
- Architecture
- Qwen3_5MoeForConditionalGeneration
- Context length (tokens)
- 262,144
- Layers
- 40
- Hidden size
- 2,048
- Attention heads
- 16
- Key/value heads
- 2
- Head dimension
- 256
- Vocabulary size
- 248,320
- Experts
- 256
- Experts active per token
- 8
- Model type
- qwen3_5_moe
Identity and Version
- Repository
- redhillsmediafl/Longleaf-35B-A3B
- Publisher
- Red Hills Media
- Task
- Text generation
- Modality
- Text
- Library
- mlx
- Parameters
- 35.1B parameters
- Languages
- en
- Revision
- dd4b62ee9c24db22879b04b10b51bdade7722a29
- First published
- 2026-10-10
- Last updated
- 2026-10-11
Files and Weights
19 files, 20.4 GB in total. The weights are 4 files totalling 20.4 GB in safetensors.
Weights4 files · 20.4 GB
Configuration8 files · 244.7 KB
Tokenizer3 files · 26.7 MB
Documentation2 files · 14.0 KB
Other1 file · 7.8 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00004.safetensors | Weights | 5.3 GB | 09f3e6ecb0b7 |
| model-00002-of-00004.safetensors | Weights | 5.4 GB | 31dcdb1c49ee |
| model-00003-of-00004.safetensors | Weights | 5.4 GB | a769b0451eaa |
| model-00004-of-00004.safetensors | Weights | 4.4 GB | 401eb0ba34e3 |
| config.json | Configuration | 23.6 KB | — |
| configuration.json | Configuration | 58 B | — |
| generation_config.json | Configuration | 202 B | — |
| model.safetensors.index.json | Configuration | 215.8 KB | — |
| preprocessor_config.json | Configuration | 390 B | — |
| processor_config.json | Configuration | 1.3 KB | — |
| release-manifest.json | Configuration | 3.0 KB | — |
| video_preprocessor_config.json | Configuration | 385 B | — |
| LICENSE | Documentation | 11.3 KB | — |
| README.md | Documentation | 2.6 KB | — |
| chat_template.jinja | Other | 7.8 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer.json | Tokenizer | 20.0 MB | 87a7830d63fc |
| tokenizer_config.json | Tokenizer | 1.1 KB | — |
| vocab.json | Tokenizer | 6.7 MB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 20.4 GB
Download from Red Hills Media
Released by Red Hills Media through its official repository on Hugging Face. Read the license.
Built From
- Derived from Qwen/Qwen3.6-35B-A3B
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 20.4 GB |
| 16-bit | 70.2 GB |
| 8-bit | 35.1 GB |
| 4-bit | 17.6 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Built on This Model
- Derived fromLongleaf-35B-A3B-Abliterated
Questions About Longleaf-35B-A3B
How much GPU memory does Longleaf-35B-A3B need?
About 84.3 GB at 16-bit and 21.1 GB at 4-bit: the weights (35.1B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Longleaf-35B-A3B on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use Longleaf-35B-A3B commercially?
Yes. Longleaf-35B-A3B is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is Longleaf-35B-A3B's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Internal research checkpoint. Routed experts average 2.0217 bits; every other weight is BF16. Weights are stored dequantized in BF16 tensors and load with stock transformers / vLLM. License follows the base model.
The abliterated version of Longleaf, our local model for coding, repository edits, debugging, and structured tools. - 35B total / 3B active parameters. - MLX 4-bit weights. - 262,144-token configured context. EvalPlus 0.3.1, 164 tasks, pass@1. Thinking off, greedy sampling, 2,048-token output cap, 262,144-token configured context. Generation used our M1 Ultra 231 / 198 / 195 tokens, respectively. CLI sample: five Terminal-Bench 2 tasks on ARM64, thinking off. Per task: 300 seconds, 24 model calls, 32 tool calls, 2,048 output tokens per call. Apache 2.0. This release contains modified model weights, based on Qwen3.6-35B-A3B, and the Abliteration changes refusal behavior, not output…
Uncensored version of Qwen/Qwen3.6-35B-A3B with refusal behavior removed via abliteration (norm-preserving orthogonalization). Zero refusals on harmful prompts. No false refusals on harmless prompts. Abliteration identifies the "refusal direction" in the model's residual stream — the linear direction that activates when the model decides to refuse — and surgically removes it from all output projection weights using norm-preserving orthogonalization. 1. Collect residual stream activations (last token position) for 512 harmful + 512 harmless prompts across all 40 layers 2. Compute mean difference vector per layer → this is the "refusal direction" candidate 3. Score layers by…
Anansi-35B-A3B balances good instruct following, narrative reasoning, and sweet prose. Specifically for my 8gb potato, because I can't run the ~30b dense models, but maybe it can be good for your potato too? Anansi was created by combining the logical trengths of two great models via a 35/65 DARE-TIES merge, followed by a targeted (LM) head tensor interpolation to graft on better prose. I think I got the balance about right for what I was aiming for. Instruct following is much better, including with more complex instructions, most of the LM-head prose improvement comes through, and the reasoning is a little longer and still narrative/character-focused without becoming excessive. The A1…
Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations This is our most spectacular outcome ever. FFT, all parameters, 16bit. 77.4 MMLU on 34b. And it talks like a dream. Although the max positional embeddings is 4k, we used rope theta of 1000000.0 and we trained with sequence length 8k. We plan to train on the upcoming 32k version as well. Our appreciation for the sponsors of Dolphin 2.9.1: - Crusoe Cloud - provided excellent on-demand 8xH100 node - OnDemand - provided inference sponsorship This model is based on Yi-1.5-34b, and is governed by apache 2.0 license. The base model has 4k context, but we used rope theta of 1000000.0 and the…