Unverified50% confidenceFactExact time
The TTT-Discover recipe cited by Reddit discussants uses Qwen3-8B with thinking mode enabled, running on two B200s.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedThe base model is Qwen3-14B, from which two versions—KGModel and CGModel—are trained separately via supervised fine-tuning (SFT).67% similarUnverifiedQwen2.5-27B是阿里云通义实验室发布的270亿参数开源模型，在MMLU、HumanEval、GSM8K等基准上接近GPT-4早期版本水平62% similarUnverifiedUnder the same Agent framework and prompt, Qwen3 27B (64K context) outperformed Leizhen T1 and Yizhentuo-Ri (both with 256K context) in PowerPoint creation.61% similarUnverifiedAfter completing the PowerPoint, Qwen3 27B called a vision model to self-review, achieving a generate–review–revise loop, ultimately producing a 16-page fully element-editable PowerPoint presentation.61% similarUnverifiedShotcut在MLT之上使用Qt框架构建图形界面，结合现代C++编写业务逻辑61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1054047API
curl https://kongchang.com/api/v1/knowledge/claims/1054047MCP
get_claim(id=1054047)