Flam's 26B MoE Falcon model returns first token in 30ms, specialized for Indic languages
testingcatalog · x · 2026-09-15
Interactive video startup Flam detailed its in-house model family: Falcon 1.0 is a 26B mixture-of-experts LLM powering Visual Agents with 30ms first-token latency, specialized for Indic languages, while Fable 2.0 is a flow-matching diffusion transformer generating 3D assets for Airboards as native alpha video from text or image prompts.
Other figures: <10ms in-video character/product/scene switching, AI compression cutting asset size 60%, 300ms first-buffer 3D streaming with no app download, and photoreal Visual Agents avatars from a single photo talking naturally in 60+ languages at 800ms latency. The family also includes identity-preserving video model Fantom 1.0 and non-autoregressive Finesse 1.0.
More from Models
- Musk: Grok 4.7 Roughly on Par with Opus 5.0, Grok 5 May Beat Everything — inductionheads · 2026-09-15
- Benchmark errors found in CritPt; GPT-5.6 hits 94.4% pass@4 after fixes — bookwormengr · 2026-09-15
- No AI is good enough for extremely high-performance software yet, says researcher on Grok — bingxu_ · 2026-09-15
- Test: 7 Models Building a T-Shirt Shop Prefer Stale Training-Data APIs Over Latest Docs — hazelcough · 2026-09-15
- Mech Interp Finding: Double Descent Is a Phase Transition Between Memorization and Generalization — gordic_aleksa · 2026-09-15
- Polymarket Prices Only 43% Odds That Grok 5 Ships by Year-End — Polymarket · 2026-09-15