dropday.ai
HomeExploreShippedLeaderboardModel Drops
Sign in

Model Drops

New capabilities, and what people make with them.

Oct 8, 2026

Celeris-1 Decision

Diffusion decision model returns calibrated probabilities plus explanations on text or images in 67 ms

Read the announcement ↗
Explore this drop ↗
▶𝕏AppsJev Fast Judgment RouterA high-efficiency AI support router that cuts LLM costs by 90% using Jev and Kimi K3.by leopardracer

Earlier drops

Oct 8, 2026Mellum2.1 Thinking12B MoE coding model post-trained with RL in real repositories for agentic edit-and-verify loops0 creations ↗Oct 7, 2026Claude Haiku 5.5Fastest cheapest Haiku for high-volume subagent classification routing and real-time tasks63 creations ↗Oct 6, 2026Mistral Large 41T-param sparse MoE for agentic coding, cyber workflows and precise visual grounding in voxel/multimodal experiments7 creations ↗Oct 5, 2026Beam501B sparse MoE open model with 23B active params for efficient coding, reasoning and agentic workloads5 creations ↗Oct 3, 2026Kolibri-178B German-English MoE with 3.46B active parameters, 1M context, and explicit reasoning modes for sovereign deployments

The most interesting things people are building with AI, all in one place — followed and tracked as they evolve from first demo to real product.

Press / to search--:--:-- ET

Get the best builds by email. No account needed.

Dropday is new — we're looking for genuine feedback.

HomeLeaderboardTop creatorsWhat is this?PrivacyTerms

A project by@gruberbuilds

HomeExploreShippedTopDrops
0 creations ↗
Oct 1, 2026MAI-Transcribe-2-StreamingReal-time streaming speech-to-text with first partials in ~100ms and top accuracy across 60 languages0 creations ↗
Sep 30, 2026Gemini 4 Argon1M-token output runs for unbroken long-horizon coding, knowledge work and cyber defense agents37 creations ↗
Sep 30, 2026Ideogram 4.5Multi-turn image edits that preserve details across dozens of iterations without accumulating artifacts or drift4 creations ↗
Sep 30, 2026Ling 3.1 FlashMoE hybrid-reasoning model with 560B total / 25B active params for extended agentic coding and tool workflows8 creations ↗
Sep 29, 2026Mercury VoiceDiffusion LLM tuned for sub-320ms first-token voice agents with tunable reasoning and tool use5 creations ↗
Sep 29, 2026GPT-6.1 SolNear-Astra agentic coding and computer-use performance at roughly one-fifth the price of Astra with 95% cache discount107 creations ↗