isyuricunha/Mythos-nano

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 5, 2026License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Mythos-nano by isyuricunha is a 3.1 billion parameter independent open model project, featuring a 32768 token context length. This model demonstrates competitive reasoning capabilities, particularly in mathematics and competitive programming, often rivaling much larger trillion-parameter systems. It is specifically optimized for complex problem-solving tasks, showing strong performance on LeetCode-style problems and mathematical benchmarks.

Loading preview...

Mythos-nano: A Compact Model for Advanced Reasoning

Mythos-nano is an independent 3.1 billion parameter language model developed by isyuricunha, designed to achieve high-level reasoning performance comparable to significantly larger models. It emphasizes verifiable feedback to enable small models to tackle frontier reasoning challenges.

Key Capabilities & Performance

  • Exceptional Mathematical Reasoning: Mythos-nano demonstrates strong performance on competitive mathematics benchmarks such as AIME, HMMT, and BruMO, often scoring within a few points of trillion-parameter systems.
  • Competitive Programming Prowess: The model excels at competitive programming problems, achieving a 96.1% aggregate pass-rate on LeetCode contests (Python), placing it among top-tier models like GPT-5.3-Codex and Gemini 3.1 Pro.
  • Instruction Following: It shows robust instruction following capabilities, as indicated by its performance on IFEval and IFBench.
  • Uncensored Nature: The model has had its refusal direction removed, meaning it will not decline requests a safety-tuned model normally would. Users are advised to use it responsibly.

Important Considerations

  • Not for Tool-Calling/Agent-Based Programming: The model was not trained on data for tool-calling or agent-based programming. It is explicitly not recommended for tasks involving function calling, API orchestration, or autonomous coding agents.
  • Reduced Safety Guardrails: Due to its uncensored nature, safety guardrails are reduced. Users are solely responsible for the outputs and legal compliance.

Good For

  • Solving competitive programming problems (e.g., LeetCode-style).
  • Tackling complex mathematical problems.
  • Use cases requiring a compact model with strong reasoning abilities.

For optimal results, recommended sampling temperatures are between 0.6 and 1.0, with support for up to 40960 output tokens for challenging problems.

Finetunes
1 models