One researcher pretrains Mirendil-Base coding model in 3.5 months for under $10M
vjkaruna · x · 2026-10-23
Mirendil announced Mirendil-Base, a coding foundation model pretrained from scratch by a single researcher in 3.5 months with under $10M in compute, aiming to drastically cut the time, expertise and compute needed for frontier-scale pretraining. Stanford's Chris Manning reacted with excitement and skepticism: can one person rival big-lab teams? Scaling results look promising.
More from Infra
- After Mozilla compromise, Rust devs eye "remediation speed" as the next hillclimb — charliermarsh · 2026-10-23
- Migrating AI workloads often means 70%+ redevelopment, warns cloud architect — DavidLinthicum · 2026-10-23
- Perplexity Computer now runs fully local on Mac — model and agent both on-device — heyneighbor · 2026-10-23
- Underdog partners with NVIDIA: Saluki 27B claims Opus 4.6 parity in under 8 GB — heyneighbor · 2026-10-23
- Google's Agent Compute Engine gives each Gemini agent a hardware-isolated micro VM — The Decoder · 2026-10-23
- The math on Meta's $825B AI capex: 20GW of compute, but DeepSeek's architecture could rewrite demand — bookwormengr · 2026-10-23