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 811 Β· π¬ 533 Β· β‘ 126.3/hr |
| π₯ | GLM-5.3 β² 1019 Β· π¬ 501 Β· β‘ 69.8/hr |
| π₯ | RustDesk β² 197 Β· π¬ 89 Β· β‘ 33/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
An FP8-quantized, 27B-parameter vision-language model for coding, professional work, research, and long-horizon agentic tasks, deployable through Transformers, vLLM, SGLang, TokenSpeed, and Qwen Cloud. Its deployment-friendly FP8 weights, image/video support, adjustable reasoning depth, and strong published coding and computer-use benchmarks make it technically notable to the Hacker News audience. β π Traction signals: Hugging Face page: 42 likes; Qwen: 96.7k followers. Benchmarks include 61.7 on SWE-bench Pro, 79.0 on QwenSWEBench, 90.3 on LiveCodeBench v6, and 84.3 on OSWorld-Verified.
β² 811 Β· π¬ 533 Β· β‘ 126.3/hr
2. GLM-5.3 β
An open-weights coding and long-horizon agent model improved entirely through scaled post-training, with added vulnerability discovery and multi-stage exploitation capabilities; weights are scheduled two weeks after launch. The August 14 launch combines large coding gains with unusually concrete cyber results and a planned open-weight release, giving Hacker News both technical depth and security implications to discuss. β π Traction signals: 50% improvement over GLM-5.2 on Z.ai Code Bench; Terminal Bench 3.0 rose from 4.6 to 28.3; DeepSWE v1.1 from 46.2 to 66.9; CyberGym scored 84.5%; 2,436 vulnerabilities identified across 269 projects, including 1,097 medium-to-high severity issues.
β² 1019 Β· π¬ 501 Β· β‘ 69.8/hr
3. RustDesk β
An open-source remote desktop that now offers unattended Wayland access, including multi-monitor support and login-screen connections after reboot, through a preview build for x86_64 Debian/Ubuntu systems. True unattended Wayland access addresses a persistent Linux remote-desktop gap. That practical breakthrough generated 197 points, 89 comments, and 33 points per hour.
β² 197 Β· π¬ 89 Β· β‘ 33/hr
4. Toast 1 β
A specialized, backend-agnostic search agent that decomposes queries, gathers and inspects evidence, and returns curated context as either a standalone retrieval agent or frontier-model subagent. Claims of frontier-level retrieval at sharply lower cost and latency, backed by detailed enterprise, legal, and search benchmarks, drove 171 points, 57 comments, and 23.8 points per hour. β π Traction signals: Up to 10Γ cheaper and 12Γ faster; 70% correctness at approximately $1.15 per task on OfficeQA Pro V2; 3.5Γ fewer tokens at the same score on 33 legal tasks; standard runs cost approximately $0.016β$0.023 per query with eight-second median latency.
β² 171 Β· π¬ 57 Β· β‘ 23.8/hr
5. AI by Hand β
Prof. Tom Yehβs educational publication on AI mathematics, algorithms, and architectures, featuring hand-worked explanations and interactive lessons and workbooks on topics such as activations, matrix multiplication, and self-attention. Its hands-on, fundamentals-first approach offers a practical alternative to high-level AI coverage. It reached 170 points and 23.5 points per hour with only 14 comments. β π Traction signals: Over 73,000 subscribers; the completed activation chapter contains 12 interactive lessons.
β² 170 Β· π¬ 14 Β· β‘ 23.5/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, and export to seven mesh or CAD formats. Lua scripting, natural CSG operators, broad exports, and BOSL2 support present a credible OpenSCAD alternative. The Show HN received 63 points, 13 comments, and 10.2 points per hour. β π Traction signals: The showcased MuSHR racecar is a 43-part assembly of 490,802 triangles.
β² 63 Β· π¬ 13 Β· β‘ 10.2/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.
