model on Hacker News

821 titles from 625 authors in the last 30 days, with 752 distinct links and 123 Show HN.

Counts titles, it does not confirm facts. Open the source before concluding anything.

Projects on this term

The score comes from our model.

• Show HN: Decision Studio – LM Studio for open-source, Jev-style decision models · score 8.3

  Bud Decision Studio is a desktop application for running open-source decision models locally, similar to LM Studio but for Jev-style 'System One' models. It provides a playground, template management, evaluation, history, API server, and on-device training with LoRA, all accessible through a polished GUI and CLI.

• Show HN: Computer use on the phone itself: a 2B model that operates Android apps · score 8.0

  Opposable is an Android app that uses a 2B-parameter GUI-Owl vision-language model to automate UI interactions via accessibility services, running entirely on-device with llama.cpp. It introduces optimizations like image token pruning, KV cache reuse, and early exit based on accessibility tree matches, plus an NPU-accelerated vision encoder. This enables private, offline computer-use agents on mobile devices.

• Show HN: Capy-flash – a 125B MoE model on two 16 GB consumer GPUs · score 8.0

  Capy-flash is a C++ inference engine that serves a 125B MoE model (Qwen3.8-Flash-Next) in NVFP4 on two 16 GB consumer GPUs by offloading routed experts to CPU RAM and using tensor parallelism over PCIe without peer-to-peer. It includes agent-oriented features like context caching, expert residency tracking, and MTP speculative decoding, with measured performance competitive with larger setups.

• Show HN: Homi, open decision model for audio, near Flash on MMSU and 50x cheaper · score 7.8

  Homi is an open decision model for audio that answers typed questions (yes/no, multiple-choice, scored levels) about audio clips, using a Qwen3-Omni-based architecture with a decision head instead of text generation. It claims strong performance on VocalGrad and MMSU benchmarks at significantly lower cost than Gemini Flash, with a fully reproducible training and evaluation pipeline.

Recent titles

1. Anthropic's Mythos-class model allegedly leaked on Hugging Face
   2026-11-01 · huggingface.co · HN discussion
2. Open Models Dashboard
   2026-11-01 · dashboard.interconnects.ai · HN discussion
3. End-to-End Learning of Hierarchical World Models for Visual Planning
   2026-11-01 · arxiv.org · HN discussion
4. Symplectic Latent Dynamics for World Models
   2026-11-01 · arxiv.org · HN discussion
5. MiniCPM-OCR 1.0: 1.3B open model scores 93.68 on Real5-OmniDocBench
   2026-11-01 · huggingface.co · HN discussion
6. Show HN: Bowerbird – Open-source NotebookLM for any model, with checked quotes
   2026-11-01 · github.com · HN discussion
7. Same model, different agent: 17 points apart on 21 MCP servers
   2026-11-01 · m3.sineframe.com · HN discussion
8. A skeleton from 20 observations beat 160 tests, a cheap model nearly caught up
   2026-11-01 · andreasrau.tech · HN discussion
9. Can my fine-tuned model pass Pangram?
   2026-11-01 · HN discussion
10. V0.5.0 of PSSA: a non-transformer language model
    2026-11-01 · github.com · HN discussion
11. Model Aero Engines Free Manufacturable Plans/Blueprints
    2026-11-01 · outerzone.co.uk · HN discussion
12. We're Training a Figma-Fluent Editing Model
    2026-11-01 · figma.com · HN discussion
13. The Elephant-Goldfish Model plugin for spec-driven AI development
    2026-11-01 · opensource.googleblog.com · HN discussion
14. Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard
    2026-10-31 · github.com · HN discussion
15. Anthropic pushes for opt-out model for Australian content
    2026-10-31 · theguardian.com · HN discussion
16. Show HN: Nvidia RE-USE speech enhancement model running in the browser on WebGPU
    2026-10-31 · re-use-demo.geronimo-labs.com · HN discussion
17. Anthropic blocks users for anything touching training larger models
    2026-10-31 · twitter.com · HN discussion
18. I post trained a model on Demoscenes
    2026-10-31 · datavorous.com · HN discussion
19. Show HN: Enclyra – 3-4B local models that do real work on an 8 GB Mac, offline
    2026-10-31 · enclyralabs.com · HN discussion
20. Show HN: A language model that lives in a boot sector
    2026-10-31 · github.com · HN discussion
21. Show HN: Tadpole – an LLM agent built from scratch then post-trained on 4B model
    2026-10-31 · github.com · HN discussion
22. Show HN: NullPrime – do AI models know when they can't know?
    2026-10-31 · github.com · HN discussion
23. Estimate and track the environmental footprint of GenAI models at inference
    2026-10-31 · ecologits.ai · HN discussion
24. Open-weight AI models can exploit the patch gap
    2026-10-31 · epoch.ai · HN discussion
25. Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories
    2026-10-31 · arxiv.org · HN discussion
26. Hi3D Launches AI 3D Model with 2048³ Voxel Resolution
    2026-10-31 · 3dprintingindustry.com · HN discussion
27. Show HN: Fastbrowse, Browser agent that picks clicks with a System One model
    2026-10-30 · fastbrowse.ai · HN discussion
28. I created a 3D model viewer that runs in the terminal
    2026-10-30 · github.com · HN discussion
29. ProgramBench: Can Language Models Rebuild Programs from Scratch?
    2026-10-30 · github.com · HN discussion
30. Show HN: Cerebellum – small model on the robot, big model for what it can't undo
    2026-10-30 · github.com · HN discussion

All trends