Unverified claim: an upcoming "Luna-level" model with 0.3B active params at $0.01/Mtok
teortaxesTex · x · 2026-11-05
- Unverified claim circulating: a "Luna-level" tiny model is coming shortly, priced at $0.01/Mtok with only 0.3B active parameters plus a terabyte-scale retrieval database.
- Description says some circuits are hardcoded ("by the Bel") for math and data movement, while the rest is trained on 100T tokens and pruned. Reposter's take: "disgusting but will be very cheap."
More from Infra
- Revolut cut AI coding inference costs ~90% by self-hosting GLM on 56 B200s instead of Claude — sebkrier · 2026-11-05
- Basket of 19 neocloud stocks (NBIS, CRWV, IREN) now merely matches the S&P 500 after wiping out 100-point lead — tengyanAI · 2026-11-05
- India plans indigenous GPU in 24 months, backs it with $15.3B semiconductor incentives — SumitGup · 2026-11-05
- OpenRouter data: DeepSeek's own API serves only 16.84% of its tokens — FinanceYF5 · 2026-11-05
- Fine-tuning a small Qwen vision model is easy — hosting it meant 10 days of idle GPU rent — allenling · 2026-11-05
- Sol 6.1 hits 24.5 tps vs promised 750 tps at $60/1M output tokens — NandaVegg · 2026-11-05