Young Tries AI

Models: guides and news

Guides, daily items and finds about models.

From the daily issues

  1. Claude Mythos access reportedly opens to individual researchers via KYC

    Partly usable
  2. tiny-voca is an 18 MB open English TTS model that runs in the browser

    Usable today
  3. Nex-AGI posts Nex-N2.6, a 321B agentic model built on GLM-5.3-Flash

    Partly usable
  4. NVIDIA tutorial serves GLM-5.3 NVFP4 across two DGX Stations with vLLM

    Not yet
  5. Claude Opus 5.5 is now the default model on paid Claude plans

    Usable today
  6. Gemini 4 Hydrogen ships in Search, the Gemini app and the API

    Usable today
  7. Edge0 open-sources its 145B and 120B models with an inference framework

    Usable today
  8. Reddit user releases Inflect v3, a family of local text-to-speech models

    Usable today
  9. Runway releases Praxis-1, an open weight world action model for robots

    Partly usable
  10. Unsloth posts updated Dynamic v3.1 GGUFs for Qwen3.8-27B on Reddit

    Usable today
  11. ElevenLabs Scribe v3 cuts word errors 27% against Scribe v2 across 102 languages

    Usable today
  12. Claude Code adds Fable 5.2 as the default Fable model with 1M context

    Usable today
  13. Anthropic releases Claude Fable 5.2, faster and cheaper than 5.1

    Usable today
  14. Anthropic shows Claude Fable 5.2 designing three working electronics prototypes

    Not yet
  15. Underdog releases a 32MB open voice model with 31 million parameters

    Usable today
  16. Mistral Large 4 and a Studio expansion target enterprise agent teams

    Partly usable
  17. Anthropic's engineering post defines agent memory and dreaming for Claude

    Not yet
  18. Phonon, a 65 MB and 75 MB on-device text-to-speech model, runs offline

    Usable today
  19. Linum releases an experimental text-to-image model trained for about $21K

    Usable today
  20. Ripple-65B Preview, a dense open-weight model with up to 1M context, is available

    Partly usable
  21. Anthropic publishes FlashProve research on proving which model produced an output

    Not yet
  22. Google rolls out Gemini 4 Argon and expands SynthID Detector

    Partly usable
  23. Reddit posts point to Qwen4 model ids in Qwen Code, with no release date

    Not yet
  24. Reddit user posts GLB-3D-Coder, a 4B model that writes Three.js scenes

    Partly usable