This living directory indexes our deep-dive engineering trackers and pricing comparisons across upcoming leaks and brand-new 2026 AI model releases.
| AI Model | Status (Nov 2026) | Key Specs / Verified Pricing | Research Breakdown |
|---|---|---|---|
DeepSeek V4.1 Flash (DS V4.1 Flash) |
Unannounced (Early vLLM PR #69107 Signal) | AMD MI300X gfx942 monokernel decode (--monokernel-layers) targeting 192 GB HBM at TP4 |
DeepSeek V4.1 Flash vLLM PR #69107 Tracker (Telegraph) |
| Thinking Machines: Inkling 2 Mini | Launched Nov 20, 2026 (Apache 2.0) | 26.4B looped MoE (~3.4B–5.4B active/token), 1M context, $0.15/1M in, $0.50/1M out, B200 vLLM PR #69104 | Running Inkling 2 Mini Locally & on vLLM (Dev.to) |
| RadixArk: DaisyRose-27B | Launched Nov 18, 2026 (Apache 2.0) | 27B dense on Qwen3.8-27B, 256K ctx, pure RL via Miles doubling Terminal-Bench 4.1 to 12.0% | RadixArk DaisyRose-27B Pure RL Deep-Dive (Paragraph) |
| Google DeepMind: Genie 4 | Live World Model | 128,000-frame rolling window (~35 min @ 60fps), $99.99/mo Google AI Ultra + Gemini API streaming | Genie 4 128K-Frame World Model Guide (Ecency/Hive) |
| Gemini 4 Neon vs. GPT-6.2 Astra | Live Flagship APIs | Neon ($3/$15, $0.15 cached, 1M ctx) vs. Astra ($8/$40, $0.40 cached, ~1.05M ctx) | 2026 Enterprise AI Model Pricing Brief (LinkedIn) |