FLock’s THIS/THAT Model Draws Korean Coverage
FLock.io’s THIS/THAT Model, released October 13 with Qwen3.5-2B and Alibaba Cloud support, is a 1.88B open-source model that returns decisions instead of generated prose. Maekyung highlighted its reported 31ms local inference and 87.8% success rate across 1,710 decision tasks.
The compelling angle is the shift from general-purpose generation to small, typed decision models for agent control paths. FLock’s claims remain vendor-reported, but the approach could make local routing, classification, and policy checks far cheaper.
- –Returns probabilities over predefined choices with zero generated output tokens.
- –Local execution improves privacy and latency for edge devices and enterprise workflows.
- –Agents can use it for routing, risk checks, tool selection, and escalation before invoking larger models.
- –Comparisons with GPT, Claude, and Jev require caution because benchmarks and calibration methods differ.
- –Alibaba Cloud’s Qwen partnership and Korean media attention give the release meaningful regional distribution potential.
DISCOVERED
1h ago
2026-10-15
PUBLISHED
1h ago
2026-10-15
RELEVANCE
AUTHOR
popmemecoin