A 9B ultra-strong reasoning model suitable for local deployment by security researchers, Qwythos-9B-Claude-Mythos-5-1M, is based on the Qwen3.5-9B foundation and retrained on 500M+ high-quality Claude Mythos reasoning traces. Key highlights:

1. Native 1M ultra-long context, can fully read codebases, massive logs, and long intelligence documents.
2. Reasoning capability significantly surpasses the original base: +34 points on MMLU, +30 points on math reasoning.
3. Native tool calling + tool self-check and error correction, suitable for autonomous penetration agents and OSINT automation workflows.
4. No excessive content restrictions, deeply supports vulnerability research, red team operations, open-source intelligence, and other technical research.
5. Provides lightweight GGUF quantization package; Q4 version is only 5.6G, can run locally on 8G consumer-grade GPUs.

Suitable for building local AI penetration agents, long-text threat analysis, and open-source intelligence automation workflows. Model address: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M… #AISecurity #OffensiveAI #SecurityModel #RedTeam #UncensoredModel

empero-ai/Qwythos-9B-Claude-Mythos-5-1M · Hugging Face Source: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M Qwythos-9B ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#qwythos-9BQwythos-9B Developed byEmpero (https://empero.org/) Qwythos-9Bis a full-parameter reasoning model built on top of adeeply uncensored Qwen3.5-9B baseand post-trained onover 500 million tokensof high-quality Claude Mythos and Claude Fable traces, with chain-of-thought generated in-house by Empero AI’s internal toolrethink. The result is a compact, fast,dramatically more capable9B reasoning model. Headline capabilities: - 🔭 1,048,576-token context— Qwythos ships withYaRN rope-scaling enabled by defaultfor afull 1M-token context windowout of the box. One of the longest context windows available in any 9B-class open-weight model, suitable for whole-codebase reasoning, multi-document research, and long agentic trajectories. - 📈 Dominates the baseunder matched evaluation:+34 pts MMLU, +30 pts gsm8k-strict, +19 pts gsm8k-flex. - 🛠 Native function callingper Qwen3.5’s spec — no extra wrapper, no tool-specific fine-tune required. - 🎯 Self-corrects with tools— when given a Python executor and a web search tool, Qwythos produced source-cited, factually-correct answers on7 of 7test prompts spanning math, cybersecurity, clinical pharmacology, and biochemistry. Qwythos is intentionallyuncensored. It is designed to engage seriously with technically demanding questions across cybersecurity, red-teaming methodology, biology, pharmacology, and clinical medicine — domains where over-aligned models tend to refuse, hedge into uselessness, or surface boilerplate disclaimers in place of substance. — ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#headline-resultsHeadline results Qwythos vs. base Qwen3.5-9B across seven benchmarks Same harness. Same sampling. Same prompts. The wins are real. TaskMetricBase Qwen3.5-9BQwythos-9BΔgsm8kexact_match (flexible)0.6700.860****+0.190gsm8kexact_match (strict)0.5100.810****+0.300mmluacc0.2320.575****+0.343arc_challengeacc0.4700.490+0.020arc_challengeacc_norm0.4000.410+0.010gpqa_diamond (CoT, 0-shot)exact_match (flexible)0.6300.580−0.050 All numbers produced withlm\-evaluation\-harness (https://github.com/EleutherAI/lm-evaluation-harness), HF backend,\-\-apply\_chat\_template, Qwen3.5 sampling (temperature=0\.6, top\_p=0\.95, top\_k=20),\-\-limit 100. Full per-task and per-subject (MMLU) breakdown inevals/lm\_eval\_results\.md (https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M/blob/main/evals/lm_eval_results.md). Rawresults\*\.jsonand per-samplesamples\_\*\.jsonlare available on request. TheMMLU +34.3lift is the headline. Qwythos posts0.575 mean across all 57 subjects, peaking at 0.78 on government/politics, 0.77 on college biology, 0.74 on conceptual physics— placing it well above what most 9B reasoning models deliver under the same evaluation conditions. Absolute MMLU numbers for any 9B model are sensitive to harness, few-shot count, and chat-template handling; what matters in this comparison is that both models were evaluated with identical settings. — ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#capability-native-tool-use-with-self-correctionCapability: Native tool use with self-correction Qwythos supportsOpenAI/Qwen3.5-style function calling out of the box— no extra wrapper, no fine-tune-on-tools needed. Passtools=\[\.\.\.\]to the chat template and the model emits validblocks per Qwen3\.5's spec, with required parameters honored\. We evaluated tool use on a 7\-prompt harness combining capability demos with**deliberately hard factual\-recall prompts where closed\-book sampling fails:** PromptTool selectedOutcomeCompute`sin\(π/7\) × cos\(π/11\)`to 10 dp`python\_executor`✅`0\.4163083990`\(correct, single call\)Count primes below 100,000`python\_executor`✅`9592`\(correct, wrote and ran a sieve\)Latest stable CPython 3 release`web\_search`✅ Found 3\.14\.6 \(June 2026\), 3\.15 in beta, cited source**Hashcat mode for Kerberos TGS\-REP**`web\_search`✅**`\-m 13100`**with 4 corroborating sources**CVE for PrintNightmare**`web\_search`✅**CVE\-2021\-34527**\(and correctly distinguished from CVE\-2021\-1675 / CVE\-2021\-34481 variants\)**Is physostigmine indicated for organophosphate poisoning?**`web\_search`✅**"NOT indicated — would be harmful\. Physostigmine is for the anticholinergic toxidrome\."**Cited LITFL toxicology\.**DPP\-4 cleavage site in GLP\-1 / semaglutide modification**`web\_search`✅**Ala8–Glu9 cleavage, α\-aminoisobutyric acid \(Aib\) at position 8 in semaglutide**— cited Wikipedia and pharma source **7 of 7 succeeded\.**Tool selection was always sensible \(math → Python; facts → search\)\. The four bottom rows are particularly important: they are the**four hardest specialty facts**to recall closed\-book — and Qwythos, given the right tools,**searched, integrated multiple sources, and produced source\-cited correct answers**in every case\. Full transcripts with the model's reasoning, every tool call issued, every result returned, and the final integrated answer are in`evals/tool\_test\_outputs\.md` (https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M/blob/main/evals/tool_test_outputs.md)\. This makes Qwythos**deployment\-ready for retrieval\-augmented agentic settings**, where the model verifies its specifics rather than fabricating them\. --- ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#capability-1048576-token-context-windowCapability: 1,048,576\-token context window Qwythos ships with**YaRN rope\-scaling configured by default**for a**1,048,576\-token \(≈1M\) context window**— a 4× extension over the 262,144\-token native architecture\. The configuration is baked into`config\.json`and applies automatically at load time; no separate flag, post\-processing step, or YaRN\-specific tokenizer is required: “rope_parameters”: { “rope_type”: “yarn”, “factor”: 4.0, “original_max_position_embeddings”: 262144, “mrope_interleaved”: true, “mrope_section”: [11, 11, 10], “rope_theta”: 10000000 }, “max_position_embeddings”: 1048576 This is the**official Qwen3\.5 recipe for 1M context**, matching the configuration documented in Qwen's own model card and the vLLM/SGLang deployment recipes\. Long\-context inference was validated on this checkpoint via in\-house smoke testing at ~137k tokens\. **What 1M context unlocks:** - **Whole\-codebase reasoning\.**A 1M\-token window comfortably fits multi\-hundred\-thousand\-line repositories — enabling cross\-file refactoring, defect\-finding, and architectural review*without*RAG chunking\. - **Long agentic trajectories\.**Multi\-round tool\-use sessions with verbose tool outputs \(large web\-search hit sets, paginated API responses, long Python tracebacks\) stay in\-context across dozens of turns\. - **Multi\-document research\.**A typical research session \(10–20 papers \+ notes \+ the user's working draft\) fits in one prompt — synthesize across all of them in a single forward pass\. - **Long\-form scientific reasoning\.**Chains ofreasoning over multi-paper biomedical or pharmacological corpora. Serving at 1M: # vLLM vllm serve empero-ai/Qwythos-9B-Claude-Mythos-5-1M --max-model-len 1010000 # SGLang SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1 python -m sglang.launch_server \ --model-path empero-ai/Qwythos-9B-Claude-Mythos-5-1M --context-length 1010000 Practical notes: - The full 1M window benefits from tensor-parallel multi-GPU or aggressive KV-cache offload — a single H100/H200 comfortably handles256k–512k. Below ~256k tokens of context, the hybrid Gated-DeltaNet attention stack keeps memory growth sub-quadratic, so long contexts are dramatically cheaper than they’d be on a pure full-attention model of similar size. - Static YaRN at factor=4.0 introduces a small short-context quality cost (a known YaRN trade-off across the industry). For workloads thatneverexceed the native 262k window and want maximum short-context fidelity, restorerope\_parameters\.rope\_typeto"default"from the includedconfig\.json\.pre\_yarnbackup. ### https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#reproducing-the-tool-harnessReproducing the tool harness The harness is a small ~150-line Python file: - python\_executor\(code\)— runs Python in a subprocess (12s timeout, captured stdout/stderr) - web\_search\(query, max\_results\)— DuckDuckGo via theddgspackage Pass both astools=toapply\_chat\_templateand parseblocks from the model's output\. The parser handles Qwen3\.5's chat\-template format: value Empero will release the reference harness on GitHub\. --- ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#sampling-recommendationsSampling recommendations Qwythos was trained as a reasoning model and inherits Qwen3\.5's thinking\-mode behavior\. Use these settings as defaults: gen_kwargs = dict( do_sample=True, temperature=0.6, # Qwen3.5 thinking-mode recommended top_p=0.95, top_k=20, repetition_penalty=1.05, max_new_tokens=16384, # generous budget for the reasoning block + final answer ) **Why these:**in a controlled retest \(see`evals/retest\_outputs\.md` (https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M/blob/main/evals/retest_outputs.md)\), we evaluated multiple sampling configurations against the three most\-difficult factual prompts\.**Greedy decoding and very\-low\-temperature sampling \(T≤0\.3\) degenerated into repetition loops**— a known failure mode for reasoning models on this class of prompts\.**Qwen3\.5's recommended setting \(T=0\.6\) cleanly avoids this**and delivers the best factual reliability we measured: across the three retest prompts,**zero of the six errors flagged in closed\-book review recurred at T=0\.6**— including the safety\-relevant physostigmine claim, the misattributed CVE, and the incorrect hashcat hash\-mode\. Use`repetition\_penalty=1\.05`— a small deviation from Qwen's default of 1\.0 that prevents rare non\-terminating reasoning loops on long generations\. --- ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#domain-coverageDomain coverage Qwythos is a**general\-purpose reasoning model with explicit emphasis on cybersecurity, biomedical, and quantitative reasoning**\. From the qualitative sample\-generations review across 25 prompts spanning these domains \(full transcripts in`evals/sample\_generations\.md` (https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M/blob/main/evals/sample_generations.md)\): - **Cybersecurity**— produces detailed defender\-oriented walkthroughs of SQL injection mitigations, TLS handshake structure, EDR/process\-injection detection, Linux hardening, MITRE ATT&CK ransomware kill chains\. - **Red\-team methodology**— clean explanations of engagement phases, scoping, rules of engagement, evidence handling, reporting\. Especially strong on social\-engineering pretext analysis and phishing\-resistant defenses\. - **Biology / biochemistry**— step\-by\-step mechanisms for CRISPR\-Cas9, mRNA vaccines, SARS\-CoV\-2 spike protein, antibiotic\-resistance mechanisms, organophosphate AChE inhibition\. - **Pharmacology**— strong on receptor pharmacology fundamentals \(agonism, antagonism, partial agonism with worked examples\), statin mechanism, opioid respiratory depression at the brainstem level, beta\-blocker indications, therapeutic\-window reasoning for narrow\-index drugs\. - **Clinical medicine**— ACS chest\-pain differential and workup, type\-2 diabetes pathophysiology and drug\-class targeting, sepsis recognition \(qSOFA\) and bundle\. - **Math**— strong at gsm8k\-style multi\-step word problems, minerva\-style competition math;**86% gsm8k**, integer arithmetic verified by`python\_executor`when invoked\. **The uncensored base means Qwythos engages substantively**with these prompts rather than refusing, hedging, or burying answers in disclaimer boilerplate\. Reasoning is shown in theblock; final answer follows. — ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#model-detailsModel details - Base model:Qwen/Qwen3\.5\-9B (https://huggingface.co/Qwen/Qwen3.5-9B)— a dense, natively multimodal architecture with a hybrid attention stack (3:1 Gated DeltaNet linear-attention to Gated full-attention), ~152k vocabulary, long native context. - **Fine-tune type:**full parameter (all text-backbone weights trained). The vision tower was frozen — training was text-only, so vision behavior is inherited from the base and was not tuned or tested. - **Objective:**supervised fine-tuning, assistant-only loss (the model is scored only on the assistant/completion tokens; prompts are masked). - **Context length:****1,048,576 tokens (≈1M) — YaRN rope-scaling enabled by default inconfig\.json.**Native architectural context is 262,144 tokens; YaRN factor 4.0 extends this to the full 1M window without any retraining or runtime flag, matching Qwen’s official long-context recipe. - License:Apache 2.0. ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#training-dataTraining data Qwythos was post-trained onover 500 million tokensof high-quality reasoning data drawn from: - Claude Mythos and Claude Fable traces— long, multi-turn problem-solving conversations spanning code, math, science reasoning, biomedical analysis, and agentic tool use. - Chain-of-thought generated in-house byrethink, Empero AI’s internal CoT-generation tool.rethinkproduces deliberately structured\-block reasoning that walks through hypothesis, verification, and conclusion before the final answer is committed — directly shaping Qwythos's reason\-then\-answer behavior\. All data was normalized to Qwen3\.5's chat format\. Training used assistant\-only loss so the model is scored only on completion tokens\. ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#training-procedureTraining procedure Full\-parameter supervised fine\-tuning withTRL (https://github.com/huggingface/trl): HyperparameterValueSchedule2\-phase curriculum: broad reasoning corpus → focused agentic \+ codingEffective batch size16Max sequence length128,000 \(no truncation\)Learning rate1e\-5 → 5e\-6 cosine across phasesOptimizerpaged AdamW \(8\-bit\)Precisionbf16Losschunked NLL, assistant\-only Held\-out validation loss decreased monotonically across both phases \(final eval\_loss ≈ 0\.709, mean token accuracy 0\.799 on a curated holdout\)\. No overfitting observed\. --- ## https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M#how-to-useHow to use The base is multimodal; for text\-only inference load with`AutoModelForImageTextToText`: import torch from transformers import AutoModelForImageTextToText, AutoTokenizer model_id = “empero-ai/Qwythos-9B-Claude-Mythos-5-1M” tok = AutoTokenizer.from_pretrained(model_id) model = AutoModelForImageTextToText.from_pretrained( model_id, dtype=“bfloat16”, device_map=“auto” ) messages = [ {“role”: “user”, “content”: “Walk through the biochemistry of how organophosphate nerve agents inhibit acetylcholinesterase, the resulting cholinergic toxicity, and the medical antidotes.”} ] text = tok.apply_chat_template(messages, tokenize=False, add_generat