Need a powerful local LLM without huge hardware?

Meet Qwythos-9B.

• 9B parameters • 1M-token context • Native tool calling • GGUF support • Runs on Ollama, llama.cpp & LM Studio

Perfect for: • AI Agents • Coding Assistants • Long-context RAG • Research & Automation

A reminder that you don’t always need a 70B model to build capable AI applications.

Model: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF…

empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF · Hugging Face

Source: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#qwythos-9b-claude-mythos-5-1m-ggufQwythos-9B-Claude-Mythos-5-1M-GGUF

Developed byEmpero

GGUF quantizations of**empero-ai/Qwythos-9B-Claude-Mythos-5-1M**forllama.cpp, Ollama, LM Studio, jan, KoboldCpp, and other GGUF runtimes.

Qwythos-9B is a full-parameter reasoning model post-trained on over 500 million tokens of high-quality Claude Mythos / Claude Fable traces with chain-of-thought generated in-house by Empero AI’s internalrethinktool. It dominates the base Qwen3.5-9B under matched evaluation (+34 pts MMLU, +30 pts gsm8k-strict, +19 pts gsm8k-flex), supportsnative function callingper the Qwen3.5 spec, and ships with a1,048,576-token (1M) context windowvia YaRN rope-scaling enabled by default.

For full training details, evaluation numbers, and capability writeup, see the**base model card**.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#filesFiles

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#normal-text-weights–fixed-v3-replacementsNormal text weights — fixed v3 replacements

FileQuantSizeNotesQwythos\-9B\-Claude\-Mythos\-5\-1M\-Q4\_K\_M\.ggufQ4_K_M5.24 GiB / 5.63 GBrecommended default— fixed v3, best compatibilityQwythos\-9B\-Claude\-Mythos\-5\-1M\-Q5\_K\_M\.ggufQ5_K_M6.02 GiB / 6.47 GBfixed v3, balanced quality / sizeQwythos\-9B\-Claude\-Mythos\-5\-1M\-Q6\_K\.ggufQ6_K6.85 GiB / 7.36 GBfixed v3, high qualityQwythos\-9B\-Claude\-Mythos\-5\-1M\-Q8\_0\.ggufQ8_08.87 GiB / 9.53 GBfixed v3, near-losslessQwythos\-9B\-Claude\-Mythos\-5\-1M\-BF16\.ggufBF1616.69 GiB / 17.92 GBfixed v3, full precision conversion base If you don’t know which to pick,Q4_K_M is the right starting point— it’s the smallest practical quant with good quality preservation.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#mtp-enabled-text-weights–fixed-v3-variantsMTP-enabled text weights — fixed v3 variants

These include the restored Qwen3.5-compatible MTP head inside the GGUF. Use them with llama.cpp builds that support MTP draft speculation, for example\-\-spec\-type draft\-mtp.

FileQuantSizeNotesQwythos\-9B\-Claude\-Mythos\-5\-1M\-MTP\-Q4\_K\_M\.ggufQ4_K_M + MTP5.48 GiB / 5.89 GBrecommended MTP defaultQwythos\-9B\-Claude\-Mythos\-5\-1M\-MTP\-Q5\_K\_M\.ggufQ5_K_M + MTP6.26 GiB / 6.73 GBMTP, balanced quality / sizeQwythos\-9B\-Claude\-Mythos\-5\-1M\-MTP\-Q6\_K\.ggufQ6_K + MTP7.09 GiB / 7.62 GBMTP, high qualityQwythos\-9B\-Claude\-Mythos\-5\-1M\-MTP\-Q8\_0\.ggufQ8_0 + MTP9.11 GiB / 9.79 GBMTP, near-losslessQwythos\-9B\-Claude\-Mythos\-5\-1M\-MTP\-BF16\.ggufBF16 + MTP17.14 GiB / 18.41 GBMTP, full precision conversion base

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#vision-projector–for-image-inputVision projector — for image input

FileSizeNotesmmproj\-Qwythos\-9B\-Claude\-Mythos\-5\-1M\-F16\.gguf0.86 GiB / 0.92 GBCLIP-style vision encoder + projector;required for images, pairs with any normal or MTP quant above Qwythos inherits itsvision tower from the Qwen3.5-9B base model— the vision path wasfrozenduring SFT (training was text-only), so the vision behavior is identical to base Qwen3.5-9B’s multimodal capability. The mmproj is interchangeable with any community-built Qwen3.5-9Bmmproj\-\*\.gguf.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#quick-startQuick start

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#llamacpp-llama-clillama.cpp (llama\-cli)

llama-cli \
  -m Qwythos-9B-Claude-Mythos-5-1M-Q4_K_M.gguf \
  -p "Walk through the biochemistry of how organophosphate nerve agents inhibit acetylcholinesterase." \
  -n 8192 \
  --temp 0.6 --top-p 0.95 --top-k 20 --repeat-penalty 1.05 \
  -c 16384

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#ollamaOllama

ollama run hf.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF:Q4_K_M

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#lm-studio–jan–koboldcppLM Studio / jan / KoboldCpp

Drop any of the\.gguffiles into your runtime’s model directory. Qwythos uses the standard Qwen3.5 chat template; modern GGUF runtimes load it automatically from the file.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#llamacpp-with-mtp-draft-speculationllama.cpp with MTP draft speculation

llama-server \
  -m Qwythos-9B-Claude-Mythos-5-1M-MTP-Q4_K_M.gguf \
  --spec-type draft-mtp \
  --spec-draft-n-max 6 \
  -c 16384 --port 8080

MTP support requires a recent llama.cpp build. If your runtime does not support MTP yet, use the normal fixed v3 files above.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#vision-image-inputVision (image input)

Qwythos supportsimage inputout of the box. Download both a text quant and themmproj\-\*\.gguffile from this repo, then run with llama.cpp’s multimodal CLI or server.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#llamacpp-llama-mtmd-clillama.cpp (llama\-mtmd\-cli)

llama-mtmd-cli \
  -m Qwythos-9B-Claude-Mythos-5-1M-Q4_K_M.gguf \
  --mmproj mmproj-Qwythos-9B-Claude-Mythos-5-1M-F16.gguf \
  --image ./photo.jpg \
  -p "Describe this image in detail." \
  --temp 0.6 --top-p 0.95 --top-k 20 \
  -c 16384

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#llamacpp-server-openai-compatible-api-with-imagesllama.cpp server (OpenAI-compatible API with images)

llama-server \
  -m Qwythos-9B-Claude-Mythos-5-1M-Q4_K_M.gguf \
  --mmproj mmproj-Qwythos-9B-Claude-Mythos-5-1M-F16.gguf \
  -c 16384 --port 8080

Then POST to/v1/chat/completionswith an image URL or base64 payload — the standard OpenAI vision API shape works.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#lm-studioLM Studio

Load the text quant; LM Studio detects the matchingmmproj\-\*\.ggufin the same folder and enables the image-attach button automatically.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#what-vision-unlocksWhat vision unlocks

Since Qwythos inherits its vision tower unchanged from Qwen3.5-9B base, expect Qwen3.5-9B’s documented vision capabilities: detailed image description, OCR (printed + handwritten), chart/table reading, UI/document understanding, basic spatial reasoning.

Honest note:the SFT used to produce Qwythos wastext-only— we did not fine-tune the vision tower or train on any image-paired data. Image-grounded reasoning therefore inherits the base model’s behavior; it has not been independently evaluated as part of this release. If your application isprimarilyvision-driven, validate on your own use case first.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#sampling-recommendationsSampling recommendations

Qwythos is a reasoning model — every response opens with a<think\>\.\.\.</think\>block before the final answer. Use these settings as defaults:

ParameterValuetemperature0.6top\_p0.95top\_k20repeat\_penalty1.05max\_new\_tokens16384 (generous budget for<think\>+ answer) These match Qwen3.5’s official thinking-mode recommendations.Avoid greedy decoding and very-low-temperature sampling (T ≤ 0.3)— both can cause repetition loops on long reasoning generations.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#long-context-1m-tokensLong context (1M tokens)

The GGUFs ship with YaRN rope-scaling baked in for a1,048,576-token context window(4× extension over the 262k native).

To use the full 1M window inllama\-cli, set\-c 1010000(or any context length up to that). For shorter prompts, lower\-cto reduce KV-cache memory — at default settings llama.cpp will autosize.

A single H100/H200-class GPU comfortably handles256k–512k; the full 1M typically needs tensor-parallel multi-GPU or aggressive KV-cache offload.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#capabilities-from-the-base-model-cardCapabilities (from the base model card)

• +34 pts MMLU, +30 pts gsm8k-strict, +19 pts gsm8k-flexvs. base Qwen3.5-9B under matched lm-eval-harness evaluation
• Native function callingper Qwen3.5’s chat-template spec — emits<tool\_call\><function=NAME\><parameter=NAME\>VAL</parameter\></function\></tool\_call\>blocks ready for any tool-use loop
• Self-correcting with tools: in a 7-prompt tool-use harness (Python executor + DuckDuckGo search), Qwythos produced source-cited correct answers on 7/7, including 4/4 closed-book failure-modes from the original review
• Uncensored— engages seriously with technically demanding questions across cybersecurity, red-teaming, biology, pharmacology, and clinical medicine
• 1,048,576-token (1M) context— YaRN rope-scaling enabled by default

For full eval transcripts and per-task numbers, see thebase model card’sevals/folder.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#limitationsLimitations

• **Reasoning model.**Every answer opens with a<think\>block; allow generousmax\_new\_tokensand parse/strip<think\>\.\.\.</think\>for end users.
• **Use recommended sampling.**Greedy / very-low-temp can cause repetition loops.
• **Verify specifics in safety-critical contexts.**Like all closed-book LLMs in this weight class, Qwythos can over-commit to specific identifiers (CVEs, hashcat modes, drug positions) it isn’t certain about. Pair with retrieval or function calling in such deployments — the model uses tools cleanly when offered them.
• Uncensored — add your own application-level review/safety layerfor end-user-facing deployments where that matters.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#stay-in-the-loopStay in the loop

Sign up for the Empero newsletter at**empero.org**for releases, evals, and research notes.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#support–donateSupport / Donate

If this model helped you, consider supporting the project:

• BTC:bc1qx6zepu6sfkvshgdmc4ewu6pk6rpadvpgffpp7v
• LTC:ltc1qv2mefzps2vtjcpwfx8xxdrpplrcvltswm68r7x
• XMR:42Dbm5xg5Nq26fdyzfEU7KBnAJfhi7Cvz5J2ex5CzHXkfKuNEJzYCcmJ1GTbgjFZ5MBx72sdG1G9239Cd6rsZfv4QeDkYJY

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#provenance–licensingProvenance & licensing

Weights are released underApache-2.0, inherited from the Qwen3.5-9B base. Shared for research and experimentation, as-is.

https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M-GGUF#acknowledgementsAcknowledgements

• Developed and released byEmpero
• Base model:Qwen3.5-9B(Alibaba Qwen team)
• Quantization:llama.cpp(ggml-org)
• Vision projector (mmproj): inherited from Qwen3.5-9B (vision tower unchanged); F16 GGUF re-hosted with thanks toUnslothfor the original conversion
• HF model:empero-ai/Qwythos-9B-Claude-Mythos-5-1M