Aloha! 🌺 Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding.
Axion-1.0-35B
Axion1-0·Instruct·released 16 Oct 2026
ollama run hf.co/mradermacher/Axion-1.0-35B-GGUF:Q4_K_MThis repository provides a reproduction checkpoint of Axion 1.0-35B, following the recipe described in Axion 1.0: Open Agent Foundation Models, Simplified. The model is designed for scientific reasoning, information seeking, and tool-augmented problem solving.
How much VRAM does Axion-1.0-35B need?
At Q4_K_M Axion-1.0-35B needs about 22 GB with an 8K context (22 GB at Q4, 70 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈22 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.
Advanced: context 8K · KV cache 16-bit
Smallest GPU for each quant and context (KV cache 16-bit)
| Quantisation | 8K | 16K | 32K | 64K | 128K | 256K |
|---|---|---|---|---|---|---|
| Q3_K_M | ||||||
| Q4_K_M | ||||||
| Q5_K_M | ||||||
| Q6_K | ||||||
| Q8_0 | ||||||
| FP16 |
≤16 GB card24–32 GB48–96 GBmore
| Quantisation | Total |
|---|---|
| Q2_KHeavy lossSmallest, noticeably worse quality | ≈14 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈18 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈22 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈25 GB |
| Q6_KNear-losslessPractically lossless | ≈29 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈37 GB |
| FP16 / BF16LosslessFull precision, as released | ≈70 GB |
Estimates for 8K tokens of context: weights + KV cache + runtime overhead. All experts stay in memory, but only the active ones are read per token. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.
How to pick a quant
Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.
- LosslessF16, Q8_0
- Near-losslessQ6_K, Q5_K_M
- Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
- Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
- Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K
Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B
How much context fits?
At 8K context, Q4_K_M needs ≈22 GB(20 GB weights + 0.6 GB KV cache + 1.5 GB overhead). Pick a GPU above to see how much context fits.
Weights + overheadKV cache
Can the Tesla P40 run Axion-1.0-35B? →
Speed vs quality per quant
Smaller quants run faster and fit in less memory, but lose quality.
Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.
Real-world speed of Axion-1.0-35B
Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →
No medians yet. Be one of the first 3 to report a setup.
All reports
Run Axion-1.0-35B locally
--quantQ4_K_Mdefault; pick your GPU to tailor itmradermacher/Axion-1.0-35B-GGUF
ollama run hf.co/mradermacher/Axion-1.0-35B-GGUF:Q4_K_M llama-server -hf mradermacher/Axion-1.0-35B-GGUF:Q4_K_MDownloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.
lms get mradermacher/Axion-1.0-35B-GGUF@Q4_K_MThen load it from the LM Studio app or with `lms load`.
vllm serve Axion1-0/Axion-1.0-35Bservices:
vllm:
image: vllm/vllm-openai:latest
command: Axion1-0/Axion-1.0-35B
ports:
- "8000:8000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 8000.
sglang serve Axion1-0/Axion-1.0-35B --host 0.0.0.0 --port 30000services:
sglang:
image: lmsysorg/sglang:latest
command: sglang serve Axion1-0/Axion-1.0-35B --host 0.0.0.0 --port 30000
ports:
- "30000:30000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 30000.
from transformers import pipeline
pipe = pipeline("text-generation", model="Axion1-0/Axion-1.0-35B", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])Quantised versions and downloads
| Repository | Format | Quant | Size |
|---|---|---|---|
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q2_K | 13 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q3_K_S | 15 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q3_K_M | 17 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q3_K_L | 18 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | IQ4_XS | 19 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q4_K_S | 20 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q4_K_M | 21 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q5_K_S | 24 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q5_K_M | 25 GB |
| mradermacher/Axion-1.0-35B-GGUF | GGUF | Q6_K | 29 GB |
About Axion-1.0-35B
This repository provides a reproduction checkpoint of Axion 1.0-35B, following the recipe described in Axion 1.0: Open Agent Foundation Models, Simplified. The model is designed for scientific reasoning, information seeking, and tool-augmented problem solving. The training recipe combines approximately 18K SFT examples with three stages: • Learn to reason: supervised fine-tuning on challenging, self-contained scientific problems constructed with model-aware feedback. • Learn to investigate: supervised fine-tuning on verifiable tasks requiring external information and tool use. • Learn through interaction: reinforcement learning over agent trajectories to refine reasoning and interaction with tools. The table compares paper-reported results with the results of the reproduction checkpoint released in this repository. All scores are percentages; higher is better. HLE denotes the full Humanity's Last Exam benchmark; BMB uses the human-difficult BioMysteryBench subset. FS denotes FrontierScience. A dash indicates an unreported result. Bold highlights the reproduction checkpoint released in this repository.
Full model card on Hugging Face →Alternatives to Axion-1.0-35B
All alternatives →NVIDIA Nemotron™ is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.
Versioning notice: This card describes the current OTel 2.0 OSFT checkpoint. The model weights may be updated over time. For reproducible evaluation or production deployment, pin a specific model revision, checkpoint hash, or release tag rather than relying on the floating latest version.
Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.
Laguna XS 2.1 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine.
Laguna XS.2 is a 33B total parameter Mixture-of-Experts model with 3B activated parameters per token designed for agentic coding and long-horizon work on a local machine. It uses Sliding Window Attention with per-head gating in 30 out of 40 layers for fast inference and low KV cache requirements.
Axion-1.0-35B FAQ
- How much VRAM does Axion-1.0-35B need?
- About 22 GB at Q4_K_M with an 8K context. Q8_0 needs about 38 GB and full precision about 70 GB. Longer contexts need more memory for the KV cache.
- Can Axion-1.0-35B run on the Tesla P40?
- Yes. It fits at Q4_K_M (about 22 GB), at an estimated 81 tokens/s. It fits comfortably on the Radeon RX 7900 XTX, Instinct MI50 32GB, RTX 5090.
- Can I use Axion-1.0-35B commercially?
- Yes. It is released under Apache 2.0. Permissive: use, modify and sell freely; keep the notice and licence text.
- What is the context length of Axion-1.0-35B?
- 262,144 tokens (256K). The full window needs far more memory than an 8K context.
- How do I run Axion-1.0-35B with Ollama or llama.cpp?
- With Ollama: ollama run hf.co/mradermacher/Axion-1.0-35B-GGUF:Q4_K_M. With llama.cpp: llama-server -hf mradermacher/Axion-1.0-35B-GGUF:Q4_K_M, which starts an OpenAI-compatible server.
Rate this model
Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.
0 comments