
We didn’t find a product being promoted in this video. Its hook, format and structure below still show why it took off.
Browse products winning on TikTokAWS's Strands Labs just open-sourced Strands Decider 2B, and it's a different kind of model. It doesn't write text. You give it some text, a question and a list of options, and it picks one, answers yes or no, or scores on a scale, with a confidence on every answer. It can never answer outside your list. Under the hood it's Qwen3.5-2B-Base with the language-model head removed and a pointer head of about a million parameters in its place. 1.9B parameters in total, one forward pass, no decoding loop. On JevBench's public set it gets 72.3% (167 of 231 tasks), 3rd of 33 models in the 2B class, and 100% of the easy tasks. The two models above it are inside the retrain noise the team itself flags, and overall it ranks 50th of 89, behind much bigger reasoning models. Speed is the point: a median 115 ms per question on an RTX 3090 and 153 ms on an M3 Pro Mac, in the team's own tests. On short tasks it had never seen, answers at 0.9 confidence or more were right about 95% of the time. Where I'd use it: tool selection, model routing, triage, guardrails and cheap evals inside an agent. Where I wouldn't: coding, chat or summaries. It can't do them. It's Apache 2.0, with the weights on Hugging Face and all the training data and scripts on GitHub. The full recipe retrains in about 11 hours on one RTX 3090. No hosted AWS API was announced. Sources: strandsagents.com/blog/introducing-strands-decider and github.com/strands-labs/strands-decider Independent briefing by CodenamePoshan, not a Strands or AWS post.
Promotes 7 products. Create free account to see more.
Outliers and viral hits: their videos on Overshot, all time.