simplescaling/s1.1-32B

Hugging Face
TEXT GENERATIONPricing:Input $2.72 / Output $4.8Concurrent Unit Cost:2Model Size:32.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Feb 8, 2025License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

The simplescaling/s1.1-32B model is a 32.8 billion parameter language model developed by simplescaling, serving as the successor to s1-32B. It features a 131,072 token context length and is specifically optimized for enhanced reasoning performance. This model leverages reasoning traces from r1, distinguishing it from its predecessor which used Gemini traces, making it particularly strong in complex problem-solving and mathematical tasks.

Loading preview...

Model Overview

The simplescaling/s1.1-32B is a 32.8 billion parameter language model developed by simplescaling, designed as the direct successor to their s1-32B model. A key differentiator for s1.1-32B is its improved reasoning capabilities, achieved by integrating reasoning traces from r1 rather than Gemini, which was used in its predecessor. This model supports a substantial context length of 131,072 tokens.

Key Enhancements & Performance

s1.1-32B demonstrates notable improvements in reasoning-intensive benchmarks compared to s1-32B:

  • AIME2025 I: Performance significantly increased from 26.7 to 60.0.
  • MATH500: Achieved 95.4%, an improvement over s1-32B's 93.0%.
  • GPQA-Diamond: Improved to 63.6% from 59.6%.

These gains highlight its strength in complex problem-solving and mathematical reasoning. The development benefited from contributions by Bespoke Labs, specifically Ryan Marten, for generating r1 traces using Curator.

Ideal Use Cases

This model is particularly well-suited for applications requiring:

  • Advanced Mathematical Reasoning: Excels in competitive math problems and complex calculations.
  • Complex Problem Solving: Its enhanced reasoning traces make it effective for tasks demanding logical deduction.
  • Research and Development: Suitable for exploring and building upon advanced reasoning capabilities in LLMs.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p