Skip to contentSkip to stories

Updated

Open source

Items with an AI score under 20 are hidden. Show low-relevance items

Oct 9

TodayOct 9Fri
  1. SGLangOfficialAI score34

    SGLang reports inference speedups for MLA, MoE, and KDA kernels

    AISGLang says restructured MLA decode kernels on Rubin, which fit a deeper pipeline in 327 KiB of shared memory, deliver a 16% speedup at batch 16 with 128K context and bit-identical output. The post also reports 20% faster full FP8 MLA at batch 1 and 20% faster KDA verify kernels after keeping weights in registers and reducing synchronization. It additionally covers fusing MoE finalization, the shared expert, 8-GPU all-reduce, and RMSNorm into one collective kernel.

  2. SGLangOfficialAI score52

    SGLang adds Rubin optimizations that speed up Kimi K3 inference

    AISGLang says it worked with NVIDIA to optimize attention, MoE, and speculative verification kernels for Kimi K3 inference on early-access Rubin hardware. It reports up to 20% faster FP8 MLA at batch 1 with 128K context, 20% faster KDA verification with bitwise-identical output, and a 5.9% end-to-end speedup from MoE tail fusion that removes 276 kernel launches per decode step. The post also says SGLang powers Miles' end-to-end RL training on Rubin, including agentic RL with 64 concurrent sandboxes on the Vera CPU.

  4. Bertholomus AIXAI score22

    DeepSeek TP=2 and TP=4 kernel kits released on GitHub

    AIA GitHub post announces that kernel kits for DeepSeek's TP=2 and TP=4 recipes are now live. The post links to a repository named deepseek-v4.1-tensorfold-tp2-2xgb10 and gives no further details on performance or features.

  5. Hacker News · AI (150+ points)BlogAI score38

    Show HN: big-arrow-on-the-screen lets AI agents draw arrows and text on macOS

    AIbig-arrow-on-the-screen (bigarrow) is a MIT-licensed macOS command-line tool and skill for Claude Code and Codex that draws arrows, boxes and text over any window. Clicks pass through, keyboard focus stays put, and each arrow removes itself after a set duration or when its agent process ends. The tool only points; it never clicks, types or captures the screen, and it requires no macOS permission to draw.

  6. dexXAI score43

    Dex Horthy posts a one-word teaser, "he cook"

    AIDex Horthy (@dexhorthy) posted only the words "he cook" on X, with no further detail. The post is a short reaction and does not describe a product, release, or result on its own. Background from the quoted post by @0xblacklight describes a serverless background agent that created a GitHub pull request from an issue.

  9. Kevin LiuXAI score32

    Jarhead, an open-source voice assistant that operates your Mac for you

    AIKevin Liu releases Jarhead, an open-source voice assistant that uses a Mac the way a person would, opening apps, clicking through Spotify, browsing, typing, running git, and marking up the screen. It runs on GPT-Live-1, which listens and talks at the same time so users can interrupt it mid-task, and its reasoning can use a ChatGPT plan with Codex, Claude Code, or an API key.

    Video from @kevskgs's post
  11. RadixArkOfficialAI score28

    Miles v0.1.2 adds Kubernetes support and torchtitan training backend

    AIRadixArk releases Miles v0.1.2, adding score centering for stable async RL and an experimental native Kubernetes backend. RL jobs can run as ordinary cluster workloads, and orchestration can restart while training continues. The release also adds torchtitan as a third training backend alongside Megatron and FSDP, plus support for DeepSeek-V4.1-Flash and MiMo-V2.6-Flash-RL.

    Image from @radixark's post
  14. Mike KnoopXAI score43

    Knoop says ARC-AGI-2 is much harder than ARC-AGI-1

    AIMike Knoop says ARC-AGI-2 is far harder than ARC-AGI-1, even amid rapid progress on math. He adds that the final open solutions will be useful artifacts to study, and notes that the Kaggle Grand Prize bonus threshold of 85% has been reached this year, the final year for ARC-AGI-2 on Kaggle.

  16. Mike KnoopXAI score62

    Tufa Labs hits 88.06% on ARC-AGI-2, clearing the Kaggle bonus threshold

    AIMike Knoop says the 85% Grand Prize bonus threshold has been reached on Kaggle. The ARC Prize 2026 leaderboard lists Tufa Labs first at 88.06%, followed by Rabbithole at 80.56% and Yi-Chia Chen at 77.22%. Knoop says this will be the final year for ARC-AGI-2 on Kaggle and expects an open-source, low-cost, offline reproducible solution and model.

  18. elvisXAI score40

    Syren Video learns your style to build AI videos from prompts

    AISyren Video, a new agentic video tool, learns preferred graphics, motion, and editing rhythm from a user's library and generates new videos from a prompt. Users refine the results through chat, and the tool is free to try in a browser or through Claude MCP, per the company's announcement. The post's author says the education sector is exploring it.

  20. ModelScopeOfficialAI score28

    Corvus-Gov-3B: a 3B Chinese government-domain dialogue model

    AIModelScope released Corvus-Gov-3B, a compact model tuned for Chinese policy Q&A, public-service consultation, and internal government or enterprise assistants. It was fine-tuned on one million Chinese government-domain dialogue samples and built on Llama 3.2 3B Instruct using LoRA SFT via LLaMA Factory. The model is released under Apache 2.0.

    Image from @ModelScope2022's post
  21. OpenBMBOfficialAI score28

    MiniCPM5-2B runs at 37 tok/s on iPhone Air

    AIOpenBMB reports that its MiniCPM5-2B model runs at 37 tokens per second on an iPhone Air. The post presents this as evidence that small open multimodal models can run on mobile devices without a cloud GPU, with NobodyWho noting the model is available in its Chat app.

  22. MarkTechPostNewsAI score44

    Underdog Releases Saluki 27B, a 2-Bit Qwen3.8-27B That Beats the Original at Tool Calling

    AIUnderdog has released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB, versus 54 GB for the full BF16 model. On Underdog Bench, Saluki scores 88 against 84 for the full model, and it raises parallel tool-call accuracy to 42 from 35. It runs on stock llama.cpp, but math and reasoning drop sharply, with AIME 2025 at 79.2 versus 96.7.

  23. Pulkit GargXAI score22

    AgentSDR launches as a free, open-source AI SDR workspace

    AIAgentSDR launches on Product Hunt as an open-source AI sales development tool under the MIT license. It combines functions that users would otherwise get from Apollo, Clay, Smartlead, Instantly, HeyReach, and a CRM in one workspace. The product uses bring-your-own API keys and charges no per-seat or per-contact fees.

    Video from @Pulkitgarg25's post
  24. vLLMOfficialAI score42

    vLLM Semantic Router team releases Decision 2.0 multi-question classification models

    AIThe vLLM Semantic Router team has released Decision 2.0, which answers multiple questions about one input in a single forward pass and outputs per-option probabilities. The post presents this as useful for routing and classification. A quoted post from Xunzhuo Liu says Decision 2.0 includes six open decision models ranging from 0.6B to 27B parameters, each topping same-size open models on the Jev Decision Index 0.3.

  25. 雷电芽衣XAI score24

    ELY Office Suite, an early Rust and GPUI native office suite, is released for macOS

    AIELY Office Suite is an open-source native office suite built with Rust and GPUI, now in an early test build for macOS, with Linux and Windows planned next week. It includes Word, PPT, Excel, PDF, Markdown, and Git client components, using LibreOffice's LOK for the office kernel, MuPDF for PDF, and MarkText for Markdown rendering. The author warns the build still has many bugs and should not be used in production.

    Video from @ZacharyZhang's post

Oct 8

Oct 8Thu
  2. meng shaoXAI score43

    Unsloth integrates Microsoft's mxc sandbox for Windows AI agent isolation

    AIUnsloth has integrated Microsoft's open-source mxc sandboxing system into Windows as an OS-level sandbox for isolating AI agent code execution. Its High mode provides real operating-system isolation that confines tool calls to specified directories, while its Low mode adds language-level checks that block dangerous commands and shell escapes. Both modes also strip secret environment variables and enforce resource limits such as 8GB memory and 600-second CPU time.

    Image from @shao__meng's post
  3. Feng XueXAI score41

    ThreatBook acquires CyberStrikeAI, a widely used AI pentesting agent

    AIThreatBook has acquired CyberStrikeAI, an open-source AI pentesting agent, after reports that attackers had used it in real intrusions. The company says it added guardrails to limit misuse but will put new capability enhancements into its commercial edition, and it argues defenders should control such tools. It also corrects an earlier threat intelligence report that linked the developer, a security engineer at Alipay, to government agencies, calling that attribution a false positive.

  4. Anil Chandra Naidu MatchaXAI score29

    Open-Voyager releases open-source creative-work agent harness

    AIOpen-Voyager is announced as a free, open-source harness for creative work, described as a Codex or Claude Code equivalent. The post says it integrates with 600+ models and links to a GitHub repository for self-hosting. The background post says the original Voyager is built for video, graphics, and games, and can drive tools such as Blender, Resolve, and After Effects.

    Video from @matchaman11's post
  5. OpenClaw🦞OfficialAI score34

    OpenClaw shares recent feature updates and upcoming roadmap plans

    AIOpenClaw says it has added many new features and quality-of-life improvements over the past few months. A video covers new models, multiplayer, interactive dashboards, memory and skills, meetings and voice, and easier Mac setup. It also previews plans for the coming months.

  6. meng shaoXAI score77

    Theo open-sources tsc-rs, a Rust port of the TypeScript 7 compiler

    AITheo, creator of the T3 Stack, open-sourced tsc-rs, a line-by-line Rust port of Microsoft's Go-native TypeScript 7 compiler, type checker, and language server under MIT, pinned to typescript-go commit 673a5f17. The author reports tsc-rs is about 1.61× faster than tsc 7 and about 2.95× faster than bun check on six real-app benchmarks on an Apple M4 Pro. The port passes all 181,711 ported Go tests, and CLI output matches the Go version on 120 open-source repos except for known edge cases such as monorepo rootDir and tsc -b incremental output.

    Why it matters: The post reports a benchmarked, test-verified Rust port of the TypeScript 7 compiler, with pinned upstream and stated edge cases useful for judging its compatibility.

    Image from @shao__meng's post
  7. Jerry LiuXAI score38

    LightOn OCR-3 now on OpenDocRouter, near Gemini 3.8 Flash at lower cost

    AILightOn OCR-3 is now available on OpenDocRouter at $0.28 per 1M input tokens and $1.40 per 1M output tokens, about $3.19 per 1k pages on ParseBench. On ParseBench, the author says it sits on the Pareto frontier for open-weight OCR models, with performance similar to Gemini 3.8 Flash low at roughly 45% lower price. It is described as decent at tables, workable for charts, and quite good at grounding.

    Image from @jerryjliu0's post
  9. 🚨 AI News | TestingCatalogXAI score62

    Atomic Agent Desktop, an open-source local AI agent app, is now available

    AIAtomic Agent Desktop is a free open-source app for macOS, Windows, and Linux that runs open models like Qwen and Gemma locally without an account. It connects to a cloud model only when selected, and its Fusion feature lets a cloud model plan a task while up to 8 local agents carry it out. The post's own text adds a setup wizard that checks RAM and suggests suitable models, and import from Claude Code, Codex, Hermes, and OpenClaw.

    Video from @testingcatalog's post
  10. AnshuXAI score38

    Pocket Aces: open-source Pokémon-Balatro roguelike game with swappable IP

    AIDeveloper Anshu Chatterjee has open-sourced Pocket Aces, a Balatro-style poker roguelike that replaces all Pokémon IP with original characters. The Pokémon assets are modular and can be swapped back in from the PokeAPI repo, though the developer notes the user proceeds at their own risk. The post says the game now works on mobile with reduced-motion options, and the developer is willing to address further bug reports.

    Video from @anshuc's post
  11. Sherwin WuXAI score60

    Harvey LAB-AA v1.1 adds hallucination gate; Grok 4.7 leads at 9.4%

    AISherwin Wu, an OpenAI employee, says the updated Harvey LAB-AA v1.1 benchmark, announced by Artificial Analysis with Harvey, is more useful than the original LAB results. The new Hallucination-Gated All-Pass Rate credits a task only when every rubric criterion passes and no material hallucination appears. Grok 4.7 (xhigh) leads at 9.4%, while GPT-6 Astra (max) at 8.6% has very few material hallucinations.

    Why it matters: The update adds a hallucination gate to a legal benchmark, showing that models with high all-pass rates can rank much lower once material errors count.

  12. elvisXAI score42

    Voyager: an open harness for creative AI work across video and games

    AIElvis Saravia argues that creative work needs domain-specific agent harnesses rather than coding-oriented ones, and he highlights Voyager as an open harness for video, graphics, and games. According to the quoted post, Voyager lets agents work with local files and drive apps such as Blender, DaVinci Resolve, and Unity, and it is designed to work with models like Opus, Astra, and DeepSeek.

    Video from @omarsar0's post
  13. Alexander DoriaXAI score46

    LightOnOCR-3 claims state-of-the-art OCR performance under 1B parameters

    AILightOn has released LightOnOCR-3, a family of OCR models in 0.8B and 4B versions that it says lead benchmarks including OlmOCR-Bench and ParseBench, with the 0.8B model positioned as the sub-1B option. The models recognize text, handwriting, images, charts and document structure in one pass, process documents up to twice as fast as LightOnOCR-2, and are released under the Apache 2.0 license.

    Image from @Dorialexander's post
  14. Dhravya ShahXAI score42

    MemoryRepo: open-source implementation of Cognition's dreaming agent memory

    AISupermemory introduces MemoryRepo.dev, an open-source implementation of Cognition's dreaming memory system built on Cloudflare Artifacts, Durable Objects, Alchemy, and Effect. The project follows Cognition's Devin memory design, which builds a memory graph across sessions and prunes stale records overnight. Supermemory says it will incorporate learnings from this research into its own product.

    Video from @DhravyaShah's post
  15. Zhihao JiaXAI score62

    Lithos AI open-sources lithos-metal for fast local inference on Apple M5 Max

    AILithos AI says it is open-sourcing lithos-metal, which uses megakernels and DSpark speculative decoding. The post claims Qwen3.8-27B reaches a peak of over 200 tokens per second per user on a single Apple M5 Max. It says users can try the tool with any coding agent in one command, and links to the code on GitHub and a technical blog.

    Video from @JiaZhihao's post