SubQ Snap decision model claims edge over Jev
A new API-only decision model claims a 100K context handled in about 660ms and a Decision Index of 61.7 versus Jev's 57.9, with no open weights.
Saw this launch announcement and feel sorry for Jev. That's because a new AI model can read 100,000 tokens and make a decision in under 1 second. SubQ Snap launched today. It's a decision model built to ingest huge amounts of context and return structured choices/probabilities. According to SubQ ⚡ 32K context → ~250 ms ⚡ 100K → ~660 ms ⚡ 384K → ~2.5 sec And it supports up to 📚 384K multimodal context 👁️ text + images 🎯 Decision Index → 61.7 🆚 Jev → 57.9 💰 $0.05 / 1M input tokens 💸 output → free It uses SubQ's Sparse Attention. Instead of normal attention getting more expensive as context grows, SubQ says its attention compute scales linearly with context length. ⚠️ No open weights. | No local download. | No model size disclosed. It's API-only for now. Link in ALT