gemini 3.5 flash is free now 😳
google quietly made their newest flash model free-tier eligible. no credit card, 1M context window, and most people are still posting about the old 2.5 models.
what you get for $0:
- gemini 3.5 flash (google's newest, free tier)
- gemini 3.1 flash-lite for high-volume cheap calls
- full 1M token context on every free model
- 1,500 requests/day, native multimodal (text, image, audio, video)
- openai-compatible, one base url change
full setup (2 min):
step 1: grab a free api key
> go to https://t.co/bvoogVXOtj
> sign in with any google account
> click "get api key", no billing setup
step 2: point any client at gemini
> works in cursor, cline, claude code, anything openai-compatible
> or just prototype in the studio ui and hit "get code"
step 3: pick your model
> gemini-3.5-flash for the strong one
> gemini-3.1-flash-lite for cheap high-volume steps
important:
- pro models left the free tier april 1, flash and flash-lite only now
- free-tier prompts can be used to train google's models, don't send sensitive data
- limits are per project, extra keys don't add quota
- 1,500 req/day resets daily, plenty for building and testing
frontier-class model + 1M context + $0
while most guides are still telling you to use last-gen models
save this, the free tier shifts almost monthly
“El deterioro de las expectativas económicas se produjo por el shock de los precios de los combustibles”: Marcel advierte por cifras de estancamiento en Chile
https://t.co/OQRXQq9v7L
ACABAN DE LIBERAR UN OCR QUE LEE 40+ PÁGINAS DE UN SOLO TIRÓN, SIN PERDER MEMORIA NI VELOCIDAD
se llama Unlimited OCR Works y resuelve el problema que tenían todos los modelos de transcripción: cuanto más largo el documento, más memoria consume y más lento se pone - este lo lee todo en una sola pasada hacia adelante
la clave es algo que llaman Reference Sliding Window Attention - mantiene una caché de key-value constante durante toda la decodificación, así que no importa si son 5 páginas o 50, el consumo de memoria no se dispara
usa un modelo de 3B parámetros pero solo activa 500M durante la inferencia - lee como uno grande y gasta cómputo como uno chico
viene con conversión de PDF integrada, batching para procesar carpetas enteras desde el inicio, y un servidor compatible con la API de OpenAI - se enchufa directo a cualquier flujo que ya tengas armado
es código abierto, está en GitHub y Hugging Face - la diferencia entre esto y los OCR que veníamos usando no es de grado, es de categoría completa
🚨 SOMEONE JUST BUILT TORLINK — A SLEEK TERMINAL-NATIVE TORRENT CLIENT THAT SEARCHES EVERY TRUSTED SOURCE AT ONCE AND DOWNLOADS STRAIGHT TO YOUR DISK.
Tired of sketchy torrent sites packed with fake buttons, popups, and dead links? torlink fixes that.
Type a query (or paste a magnet), and it hits a curated list of trusted sources in parallel — FitGirl (games), YTS, 1337x, EZTV, Nyaa, SubsPlease & more. Results stream in live with sizes and seed counts. Arrow + d to download.
Highlights:
• One command: npx torlnk (just needs Node)
• Background queuing + resume support
• Auto-seeds by default (easy toggle)
• Clean keyboard TUI — nothing leaves your machine except torrent traffic
Perfect for grabbing games, movies, TV, or anime without the usual headache. MIT open source, zero setup, and stupidly simple.
This is the terminal tool power users have been waiting for.
Who’s trying this on their next download? 🔥
imagine you could run a Claude-level reasoner locally smoothly on RTX 4060 8GB laptop?
Qwythos-9B by Empero delivers serious reasoning gains with 1M context and tool calling.
Qwythos-9B-Claude-Mythos-5 post-trained on 500M+ high-quality Mythos/Fable CoT traces.
- 500M+ Claude Mythos reasoning tokens
- 1 Million token context
- Tool use & agents that actually work
- Native function calling
- Runs fully local
- Q4 only 5.2GB
- https://t.co/733WWdVGtg
🇻🇪 | DESASTRE EN VENEZUELA: Nuevo vídeo de los terremotos del 24 de junio que muestra el grado de fuerza de los movimientos de la tierra mientras de fondo los edificios colapsaban.