Reflection Llama-3.1 70B

Model created by Matt Shumer which is trained with a new technique called Reflection-Tuning that teaches a LLM to detect mistakes in its reasoning and correct course. The model was trained on synthetic data."

See also:

• UPDATE: Interesting the Reflection 70b seems to be not quite what it claims Promptfondler drama: Shumer’s ‘no-hallucination’ Reflection 70B AI turns out to be two other models in a trenchcoat

• LinkedIn: Matt's post announcing this model

• Medium Reflection 70B: Enhancing Open-Source AI with Self-Correcting Abilities
• You Tube New LLM DESTROYS Every Other Model with "Self Healing" (Open Source)
• YouTube Live Chat with Matt Shumer about Reflection 70b!
• Reddit: Reflection 70B: Hype?
• Hugging face Model weights

These models are currently available via Open Router in two versions (a paid and a free version)

Free version

mattshumer/reflection-70b:free | Updated Sep 6 131,072 context | $0/M input tokens | $0/M output tokens

Paid version

mattshumer/reflection-70b | Updated Sep 6 | 131,072 context | $0.35/M input tokens | $0.4/M output tokens