FrontierAugust 8, 2026via MarkTechPost
Pokee AI Releases Pokee-Isaac 28B: A 10M-Token Context Agentic Model Built to Run Inside the Customer Boundary
Why it matters
A capable 28B model with extreme context length (10M tokens) designed for on-premises and customer-boundary deployment. The benchmark gap at ultra-long context is substantial; this addresses a real gap for enterprises that can't send data to third-party APIs. Licensing model (not open-weight) and pricing signal a different commercialization path than mainstream open models.
Key signals
- Pokee-Isaac 28B: 28B parameters, 10M token context window
- RULER benchmark: 93.3% at 10M tokens (baselines return 0.0 beyond 2M tokens)
- BFCL v4: 70.94 (first place)
- Terminal-Bench 2.1: second place
- Prefill: 137,200 tokens/s at full context on single B200
- Decode: ~335 tokens/s
- Deployment: VPC, on-premises, on-device only (weights not published)
- Pricing: $0.15/$1.00 per million tokens (prefill/decode)
- Category: agentic model
The hook
93.3% on long-context evals where baselines hit 0.0. Pokee-Isaac 28B ships with 10M tokens—and won't leave your VPC.
Pokee AI released Pokee-Isaac 28B, a 28B text-only foundation model with a 10M-token context window built to run inside the customer boundary. It scores 93.3% on RULER at 10M tokens, where every baseline in its comparison panel returns 0.0 beyond 2M, and leads BFCL v4 at 70.94 while placing second o…