KU-AGI/RetroReasoner-RoundTrip-8B

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

KU-AGI/RetroReasoner-RoundTrip-8B is an 8 billion parameter forward reaction prediction model, fine-tuned from Qwen/Qwen3-8B, designed for round-trip evaluation in the RetroReasoner project. This model predicts the product of a chemical reaction given the starting materials in SMILES format. It specializes in direct chemical reaction outcome prediction, distinguishing it from retrosynthesis models, and operates with a context length of 32768 tokens.

Loading preview...

RetroReasoner-RoundTrip-8B Overview

KU-AGI/RetroReasoner-RoundTrip-8B is an 8 billion parameter language model developed by KU-AGI, specifically fine-tuned from Qwen/Qwen3-8B. Its primary function is forward reaction prediction, meaning it takes chemical reactants (provided as SMILES strings) and predicts the resulting product of the chemical reaction. This model is a component of the larger RetroReasoner project, where it is utilized for round-trip evaluation.

Key Capabilities

  • Forward Reaction Prediction: Predicts chemical products from given reactants.
  • SMILES Input: Processes starting materials provided in SMILES format.
  • Specialized Fine-tuning: Optimized for chemical reaction outcomes, not retrosynthesis (for retrosynthesis, refer to KU-AGI/RetroReasoner-RL).
  • Qwen3 Chat Template: Designed to work with the Qwen3 chat template for prompt formatting.
  • Direct Answers: Generates short, direct completions without extensive chain-of-thought.

Good For

  • Chemical Reaction Outcome Prediction: Ideal for scenarios requiring the prediction of products from known reactants.
  • Research in Chemical AI: Useful for researchers working on chemical synthesis and reaction modeling, particularly within the RetroReasoner framework.
  • Round-Trip Evaluation: Specifically designed for its role in evaluating retrosynthesis models by predicting forward reactions.