Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Recursive AI Releases Ripple-65B Open-Weight Model Scoring 80.52 on TerminalBench

3 sources · 3 articlesFirst reported Updated ()1
What's new

November 6: The Decoder added pretrained on 12 trillion tokens at $32 million cost, 79.1% on SWE-Bench Pro v2.0 Hard, Apache 2.0 on Hugging Face, and Nvidia optimization.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Recursive AI released Ripple-65B, a 65B dense open-weight model scoring 80.52 on TerminalBench 2.1 using custom synthetic data and RL recipes.

Recursive AI trained Ripple-65B Preview, a 65 billion parameter dense open-weight model optimized for agentic coding. The model was pretrained on 12 trillion tokens at a development cost of $32 million. It achieves 80.52 on TerminalBench 2.1 and 79.1% on SWE-Bench Pro v2.0 Hard, available on Hugging Face under Apache 2.0 license. It uses custom synthetic data pipelines and a mixed RL recipe to reach the score with 1.4 million rollouts, 70 times fewer than prior SOTA.

Key facts
  • 65B dense Transformer parameters
  • 80.52 on TerminalBench 2.1
  • 1.4M rollouts 70x fewer than prior SOTA
  • Pretrained on 12T tokens at $32M cost
  • Apache 2.0 license on Hugging Face
  • 79.1% on SWE-Bench Pro v2.0 Hard
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Nov 4, 2026Hacker News previewed Ripple-65B using custom synthetic data and RL recipes reaching 80.5 on TerminalBench · Hacker News
  2. Nov 4, 2026MarkTechPost reported release of 65B open-weight model scoring 80.5 on TerminalBench 2.1 · MarkTechPost
  3. Nov 6, 2026The Decoder reported additional details on 12T token pretraining at $32M cost and SWE-Bench score · The Decoder

What each outlet adds

beyond the shared facts
  • Hacker NewsCustom synthetic data pipelines, quality filtering, RL recipes with truncated importance sampling, and 1.4M rollouts 70x fewer than prior SOTA
  • MarkTechPost65 billion parameters and TerminalBench 2.1 score of 80.5
  • The DecoderPretrained on 12 trillion tokens costing $32 million, 79.1% on SWE-Bench Pro v2.0 Hard, Apache 2.0 license on Hugging Face, and Nvidia hardware optimization
Vendors
Ricursive IntelligenceRicursive AI
AI models
Ripple-65BRipple-65B Preview

Coverage

oldest first · one row per article · UTC
  1. · 4d ago Hacker News · AIFirst report

    Ripple-65B Preview Achieves U.S. Open-Weight SOTA on Agentic Coding

    The model uses custom synthetic data and RL recipes to reach 80.5 TerminalBench with 1.4M rollouts, 70x fewer than prior SOTA.

  2. · 4d ago MarkTechPost

    Ricursive AI releases Ripple-65B preview scoring 80.5 on Terminal-Bench 2.1

    Ricursive AI releases Ripple-65B, a 65B dense open-weight model scoring 80.5 on Terminal-Bench 2.1.

  3. · 2d ago The Decoder

    Ricursive releases Ripple-65B Preview open model

    Pretrained on 12 trillion tokens at a cost of $32 million, the model scores 80.52 on TerminalBench 2.1.