Hottest service today: Qwen 3.8 27B
HN TRACTION RADAR
π 6 similar services trending on HN today
Service traction analysis, handpicked from the Hacker News front page
π 6 analyzedπ‘ 30 in candidate pool
We picked the products and services that drew the hottest response on the Hacker News front page today, read each service page ourselves, and analyzed what they do and why they took off.
Today's Top 3 by Traction
| π₯ | Qwen 3.8 27B β² 796 Β· π¬ 520 Β· β‘ 128.9/hr |
| π₯ | GLM-5.3 β² 1016 Β· π¬ 501 Β· β‘ 70.9/hr |
| π₯ | RustDesk β² 196 Β· π¬ 88 Β· β‘ 34.4/hr |
| π Traction Analysis | 6 |
π Click a card to go to that service's page.
π Traction Analysis6
1. Qwen 3.8 27B β
A 27B-parameter native vision-language model for coding, research, professional work, and long-horizon agents, distributed as FP8 weights compatible with Transformers, vLLM, SGLang, and TokenSpeed. A deployment-friendly open multimodal model drew attention through strong published coding and computer-use results, FP8 performance described as nearly identical to the original, and controllable reasoning depth. β π Traction signals: Hugging Face shows 42 likes and 96.7k Qwen followers. Reported scores include 61.7 on SWE-bench Pro, 79.0 on QwenSWEBench, 90.3 on LiveCodeBench v6, and 84.3 on OSWorld-Verified.
β² 796 Β· π¬ 520 Β· β‘ 128.9/hr
2. GLM-5.3 β
An open-weights coding and agentic model built from GLM-5.2βs base model through scaled post-training, targeting complex coding, long-horizon professional tasks, and vulnerability discovery and exploitation. The release combines substantial coding gains with unexpectedly fast-growing cyber capabilities and a commitment to release weights after two weeks of safety evaluation, making the claims technically consequential. β π Traction signals: Z.ai reports a 50% improvement on Z.ai Code Bench; Terminal Bench 3.0 rose from 4.6 to 28.3 and DeepSWE v1.1 from 46.2 to 66.9. It tracked 2,436 vulnerabilities across 269 projects, including 1,097 critical and high-severity findings; 53 were publicly disclosed and 2,383 under embargo.
β² 1016 Β· π¬ 501 Β· β‘ 70.9/hr
3. RustDesk β
RustDeskβs preview build enables unattended remote access on Wayland, including multi-monitor sessions and login-screen connections after reboot, initially for x86_64 Debian/Ubuntu systems. True unattended Wayland access addresses a persistent Linux remote-desktop gap, especially versus products still requiring Xorg or offering experimental support. It earned 196 points and 88 comments at 34.4 points per hour.
β² 196 Β· π¬ 88 Β· β‘ 34.4/hr
4. Toast 1 β
A specialized, backend-agnostic search agent that decomposes queries, gathers and inspects evidence, then returns curated context either standalone or as a subagent for frontier models. Claims of frontier-level retrieval at sharply lower cost and latency, backed by financial and legal benchmark results, drove 169 points and 57 comments at 24.5 points per hour. β π Traction signals: Mixedbread reports up to 10Γ lower cost and 12Γ higher speed. GPT-5.6 Sol with Toast 1 reached 70% on OfficeQA Pro V2 at approximately $1.15 per task. On 33 legal tasks, tokens fell from 80.6M to 23M at the same score of 55; a standard run costs approximately $0.016β$0.023 with eight-second median latency.
β² 169 Β· π¬ 57 Β· β‘ 24.5/hr
5. AI by Hand β
Prof. Tom Yehβs publication teaches AI mathematics, algorithms, and architectures through articles, interactive lessons, workbooks, and visual explanations covering topics such as activations and self-attention. Its hands-on, visual approach to understanding AI fundamentals resonated strongly despite limited discussion, reaching 165 points with 14 comments at 23.9 points per hour. β π Traction signals: Over 73,000 subscribers; the Activation chapter contains 12 interactive lessons.
β² 165 Β· π¬ 14 Β· β‘ 23.9/hr
6. LuaCAD β
A scriptable parametric 2D/3D CAD system embedding Lua 5.4 in a Rust engine, with direct CSG evaluation, live preview, OpenSCAD generation, and multiple mesh export formats. Lua scripting, natural CSG operators, native BOSL2 support, and broad export options offer a practical OpenSCAD alternative. The Show HN received 62 points and 13 comments at 10.6 points per hour. β π Traction signals: The showcased MuSHR racecar is a 43-part assembly of 490,802 triangles.
β² 62 Β· π¬ 13 Β· β‘ 10.6/hr
The traction index is calculated from actual Hacker News upvote and comment activity (velocity per hour). The analysis covers the Hacker News front page.
