reaperdoesntknow/Dualmind-Qwen-1.7B-Thinking

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Mar 30, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Dualmind-Qwen-1.7B-Thinking is a 2 billion parameter Qwen3ForCausalLM developed by Convergent Intelligence LLC: Research Division, fine-tuned using the DualMind SFT methodology. This model specializes in extended deliberation and self-correction, having been trained on over 2.5 million tokens of Claude Opus 4.6 reasoning traces. It excels at absorbing complex reasoning patterns, including backtracking and nuanced self-correction, making it suitable for tasks requiring deep, multi-phase thought processes.

Loading preview...

Dualmind-Qwen-1.7B-Thinking: Opus Reasoning Variant

This model, developed by Convergent Intelligence LLC: Research Division, is a 1.7B parameter Qwen3-based model specifically trained to emulate the reasoning patterns of Claude Opus 4.6. Utilizing the DualMind SFT methodology, it was fine-tuned on over 2.5 million tokens from the Opus-4.6-Reasoning-3000x-filtered dataset, which contains curated reasoning chains from Anthropic's frontier model with refusals removed.

Key Capabilities & Differentiators

  • Opus-like Deliberation: Absorbs the nuanced self-correction, backtracking, and synthesis characteristic of Claude Opus 4.6, reflecting genuine uncertainty navigation rather than simple pattern completion.
  • DualMind SFT: Implements the DualMind thesis, where the cognitive loop (explore → examine → respond) emerges from the training signal of a teacher model exhibiting multi-phase reasoning.
  • Strong Foundation: Built upon Disctil-Qwen3-1.7B, a DISC-refined model from the DistilQwen distillation chain, providing a robust structural base.
  • Extended Reasoning: Designed to produce longer, more deliberative reasoning chains, with generation tips suggesting max_new_tokens up to 2048.

Ideal Use Cases

  • Complex Problem Solving: Suited for tasks requiring deep thought, self-correction, and multi-step reasoning.
  • Cognitive Simulation: Useful for applications where emulating human-like deliberation and uncertainty handling is beneficial.
  • Research in Reasoning: Provides a compact model for studying the emergence of complex reasoning from teacher-student distillation, particularly the "shape of deliberation."

Finetunes
1 models