squ11z1/Mythos-nano
Mythos-nano by squ11z1 is an independent 3 billion parameter open model project, not an official Anthropic release. This model demonstrates strong performance in competitive mathematics and coding problems, achieving scores comparable to much larger models on benchmarks like AIME, HMMT, and LeetCode. It is specifically optimized for reasoning tasks, particularly in competitive programming, and is noted for its uncensored nature with reduced safety guardrails.
Loading preview...
Mythos-nano: A Compact Reasoning Powerhouse
Mythos-nano, developed by squ11z1, is a 3 billion parameter independent open model project that challenges the notion that only large models can achieve frontier reasoning capabilities. It is specifically designed and optimized for competitive mathematics and coding problems, demonstrating performance on par with trillion-parameter systems in these domains.
Key Capabilities & Performance
- Exceptional Reasoning: Achieves high scores on competitive math benchmarks (e.g., AIME, HMMT, BruMO) and competitive programming platforms (LeetCode), often within a few percentage points of significantly larger models.
- Competitive Programming: Excels at LeetCode-style problems, with a 96.1% pass rate, outperforming many larger models.
- Uncensored Output: The model has had its refusal direction removed, meaning it will not decline requests that a safety-tuned model normally would. Users are responsible for its outputs and legal compliance.
When to Use Mythos-nano
- Competitive Programming: Ideal for solving complex algorithmic and data structure problems.
- Mathematical Reasoning: Strong performance in advanced mathematical problem-solving.
- Research & Experimentation: Suitable for exploring the capabilities of smaller models in reasoning tasks, especially where an uncensored model is desired for specific research purposes.
Important Considerations
- Not for Tool-Calling: The model was not trained on tool-calling or agent-based programming data, making it unsuitable for tasks involving function calling, API orchestration, or autonomous coding agents.
- Reduced Safety Guardrails: Due to its uncensored nature, users must exercise caution and responsibility, as safety guardrails are significantly reduced.