Ripple-65B claims open-weight agentic coding SOTA trained from scratch for just $32M
schwentker · x · 2026-11-05
RicursiveAI released Ripple-65B Preview, a dense 65B LLM pretrained entirely from scratch, claiming a new US open-weight SOTA on agentic coding that beats much larger models.
- Cost: the team says full R&D — ladder development, data mixing, evals, synthetic data, post-training and the final hero run — totaled $32M, about one-tenth of the previous open-weight SOTA's cost.
- Method (screenshot): starting from Olmo 3's model ladder, iterative small-scale interventions (StepFun scaling laws, pre+post norm, sigmoid gated attention, global NoPE + LR cooldown to zero, value residuals, code-upsampled custom dataset) compounded into up to 16x compute efficiency gains on code CE at 4B scale — matching Olmo 3 code performance with 16x less compute, with the gap growing at scale.
- swyx amplified the release, calling it "the Avengers of open model SOTA." Third-party verification is still lacking; figures are official claims.
More from Infra
- Claim: CoreWeave Cut Off From Bond Markets, Unable to Meet 2027 Obligations — SumitGup · 2026-11-05
- Bratton: "Against Data Centers" Is the "No Nukes" of the 2020s — and Will Backfire — bratton · 2026-11-05
- Cornell sets semiconductor bandgap record with 7+ eV doped aluminum gallium oxide devices — AviatorAshley · 2026-11-05
- SiTime revenue hits $298.7M, up 257% YoY — BenBajarin · 2026-11-05
- George Hotz to teach CMU course where students build an ML compiler from scratch up to training LLMs on GPUs — QuixiAI · 2026-11-05
- CI costs scale directly with token prices, dev observes — aniketmaurya · 2026-11-05