LOADING DATASHEET
LOADING DATASHEET
DAY 1 ON THE BOARD
#235 of 349 assistants, 315 voices so far
by Moondream
JUST LANDED · UPDATING HOURLY
RANKED #235 OF 349 ASSISTANTS · OVERALL #363 OF 541 RANKED · VIBE SCORE 5.5 · 315 VOICES
COPY BADGE pastes a README snippet with the rank SVG.
Moondream 3.1 is a vision-language model from Moondream, the successor to Moondream 3 Preview, with 9B total parameters and 2B active (mixture-of-experts). It handles visual question answering, captioning, OCR, and...
3 mentions
2 mentions
26 weeks · 318 voices
TOO QUIET
Not enough weekly voices to call a trend yet.
Open weights. You can use a hosted service, or download it and run it yourself, free.
People are mostly frustrated with Moondream 3.1 right now. Biggest gripe: getting facts right.
30 POSTS · 285 COMMENTS · REDDIT · GITHUB
4 thumbs up · 6 thumbs down
0 thumbs up · 8 thumbs down
1 thumbs up · 3 thumbs down
2 thumbs up · 2 thumbs down
Other models the crowd has fully reviewed, starting with multimodal models like this one.
I am trying something similar as MoonDream, and Molmo. i.e make the language model capable of producing normalized coordinates of objects asked about. "Point: Dog" e.g. I am trying to make smolvlm do this as a fun project to get better understanding. I am…
Moondream3 and Salesforce GTA-1 for UI grounding in computer-use agents
Investors temper Anthropic IPO valuation expectations over revenue growth slowdown, sources say
ARCHITECTURE: McCulloch's Neuron: Each LLM call now uses explicit num_ctx: 2048 per-request, forcing a clean KV cache every turn. The LLM is genuinely stateless — born, perceives, responds, releases. Memory Ring is the sole source of continuity. The model is…
So what are they going to call super intelligence then?
I have been summoning him for years. Idc what Prickly Thiel says.
So no thought of a separate backup? That’s so ignorant.
Customize your AI to help map ALL OF EARTH'S NATURAL PROCESSES and look for novel connections & hypotheses. Yes - you can make a scientific breakthrough with your chatgpt subscription and a little bit of ingenuity.
Author disclosure: these are my independent projects. I'm sharing two Mamba2-8B research releases exploring different compression tradeoffs, each with a trained Resurface-style recall adapter. These are separate configurations; their memory numbers should not…
sample text
tbh i've been thinking about this trajectory a lot lately. right now we're in the phase where AI is crushing technical friction, forcing devs and analysts to step up and act like project managers just to keep their seats. but that’s just a temporary pit stop.…
- opened automatically by the GitHub Action. this branch carries 9 price rows and 0 removals. new rows source model observed input (/1M) cache read (/1M) cache write 5m/1h (/1M) output (/1