orangejuicesmith/Mythos-Nano

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 9, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

Mythos-nano by orangejuicesmith is a 3.1 billion parameter independent open model project, notable for its strong performance in competitive mathematics and coding tasks despite its small size. It demonstrates frontier reasoning capabilities, achieving high scores on benchmarks like AIME and LeetCode, often comparable to much larger models. This model is optimized for complex problem-solving, particularly in programming and mathematical domains, and is released under the MIT license.

Loading preview...

Mythos-nano: A Compact Model for Advanced Reasoning

Mythos-nano, developed by orangejuicesmith, is an independent 3.1 billion parameter language model project that stands out for its exceptional reasoning capabilities, particularly in mathematics and competitive programming. Despite its small size, it achieves benchmark scores often comparable to trillion-parameter systems, demonstrating that verifiable feedback can enable small models to reach frontier reasoning performance.

Key Capabilities

  • Competitive Mathematics: Achieves high scores on benchmarks like AIME25 (91.4%), AIME26 (94.3%), HMMT25 (89.3%), and BruMO25 (93.8%). With CLR (96.7% AIME25, 97.1% AIME26).
  • Competitive Programming: Scores 96.1% (123/128) on LeetCode contests (Python pass-rate), placing it near top-tier models like Gemini 3.1 Pro and GPT-5.3-Codex.
  • Abliterated (Uncensored): The model has reduced safety guardrails, meaning it will not decline requests a safety-tuned model normally would. Users are responsible for its outputs.

Good for

  • Competitive Programming Problems: Specifically recommended for tasks like LeetCode-style challenges.
  • Mathematical Problem Solving: Excels in complex mathematical reasoning.
  • Research and Experimentation: Its compact size and strong reasoning make it suitable for exploring advanced AI capabilities on more constrained hardware.

Important Considerations

  • Not for Tool-Calling/Agent-Based Programming: The model was not trained on data for function calling, API orchestration, or autonomous coding agents.
  • Reduced Safety Guardrails: Due to its uncensored nature, users must exercise caution and responsibility, as the model will not refuse requests based on safety.

Mythos-nano is available under the MIT License, with GGUF formats provided for llama.cpp and Ollama.

Finetunes
1 models