Pokee Launches 28B Model: 10M Token Context as Iteration Memory, Single-GPU Deployment
Kyrannio · x · 2026-08-07
Pokee released the Pokee-Isaac 28B model, claiming to be the world's first frontier agentic model with a real 10M-token context window, deployable on a single GPU (starting from RTX 4090).
Using a non-decoder-only architecture, it achieves 93.3% on RULER at 10M tokens and up to 137K tokens/s prefill on a single B200. A developer demonstrated its practical application: using the massive context as "iteration memory", allowing the model to patch its own code in-place continuously and successfully build a playable 3D FPS game in a single HTML file.
Pricing is set at $0.15/M input and $1/M output, with support for local and on-device deployment.
More from coding & agent
- Dev Bypasses Claude Safety Filters by Banning Hundreds of Trigger Words via Hooks — chaumian · 2026-08-07
- Cloudflare Slashes Astro Open Issues by 85% Using Isolated AI Subagents — JeremyCMorgan · 2026-08-07
- a16z's Yoko Li on Making AI Agent Loops Converge and Know When to Stop — stuffyokodraws · 2026-08-07
- Kestrel: Open-Source Local Engineering Agent with Layered Memory — John_Miracleworker · 2026-08-07
- Developers discuss ideal AI companions: traceable memory and lasting relationships — ZeroStateReflex · 2026-08-07
- The Era of Model-Harness Co-Design in Applied AI — ypatil125 · 2026-08-07