Alexa Benchmark retweeted
best run on Visual Cipher Bench averages about 10.7 of the 40 puzzles. most of what kills the agents is working out the rule in the first place
🤝 Paid partnership
Coding agents can now grind through long-horizon tasks, as long as they get a written spec and a test suite.
What happens if the spec is a drawing, and nothing tells if they're right?
We gave 8 frontier models 40 visual puzzles. The best configuration solved 26.7%. Three models solved none.
Introducing Visual Cipher Bench from Einsia AI, a benchmark for long-horizon visual reasoning. Agents get tools, then after 60 minutes, no written rules and no correctness feedback is given.
40 original puzzles · 67 drawings · 8 models · 29 configurations · 3,480 graded attempts
🌐Project: lab.einsia.ai/visual-cipher-…
💻Github: github.com/Einsia/Visual-Cip…
📄Paper: arxiv.org/abs/2611.16861
Alexa Benchmark retweeted
This moment. Someone asks about a number from an old meeting, one long press, and what you covered last month is on the card.
🤝 Paid partnership
Alexa Benchmark retweeted
I watched an AI debug its own game from screenshots.
Now Tencent has released a version of that AI built to make everyday tasks faster and more cost-efficient.
Let me explain.
This cyberpunk racer was coded from scratch in about 4 minutes of generation using Hy4.
The first build crashed after 2 seconds.
Two rounds of screenshots later, the model had found the bugs on its own and the game ran clean.
That was already impressive. But Tencent's new Hy4 flash might be even more interesting for everyday use.
It supports multimodal inputs, a massive 1-million-token context window, and can handle everything from document analysis and research to presentations, spreadsheets and simple coding tasks.
Tencent tested Hy4 flash against DeepSeek-V4.1-Flash-1017 across everyday productivity tasks.
The results?
→ Hy4 flash completed tasks faster.
→ It cost approximately US$0.04 per task versus US$0.18 for DeepSeek in Tencent's blind test.
→ That's nearly 80% lower cost per task.
On Tencent's WorkIndex benchmark, Hy4 flash's per-task cost was approximately one-eighth that of DeepSeek-V4.1-Flash-1017, with DeepSeek set to the Max tier using peak pricing.
The interesting part is that Tencent isn't trying to replace its flagship model.
Hy4 flash is designed for the everyday stuff: research, documents, slides, data analysis and simple coding.
The full Hy4 model is still the better choice for complex reasoning, advanced software engineering and ambitious projects like building an entire game from scratch.
Think about it.
Why pay for maximum intelligence on every single task when a faster, cheaper model can handle most of your daily workload?
And the best part?
Hy4 flash is completely FREE on WorkBuddy until November 30.
Try it yourself:
workbuddy.ai/
@TencentHunyuan @TencentAI_News @WorkBuddy_AI
Alexa Benchmark retweeted
When a voice agent mishears a drug name, someone has to call the patient back.
ElevenLabs just shipped Scribe v3 Live Medical, real-time transcription built for medical vocabulary.
Honestly this is the part of healthcare AI I want to see get better.
🤝 Paid partnership
Scribe v3 Live Medical brings medical transcription to live calls.
It helps voice agents get medications, procedures, and symptoms right, and teams can add up to 10,000 of their own terms, like provider and drug names.
Now in ElevenAgents.
elevenlabs.io/blog/scribe-v3…
Alexa Benchmark retweeted
Dashcam versus dashboard is how I'd sort AI meeting gear. Pins and pendants replay what happened. Verso wants to be the thing you glance at while it's happening.
🤝 Paid partnership
Alexa Benchmark retweeted
If you use ChatGPT or Claude for work, you know the tax: three paragraphs of background before the first useful answer. Zac's @flowtica_ai video makes a good 3 minute case for capturing that context where it happens.
🤝 Paid partnership
Alexa Benchmark retweeted
Le pedí a una IA 4 versiones de mi anuncio para un A/B test y en una de ellas se negó a mentir 😳
Tenía un anuncio UGC de una crema con SPF 30 y quería saber qué gancho vende más en Meta esta semana. En menos de 2 minutos tenía 4 anuncios con la misma creadora, el mismo bote y el mismo botón. Lo único que cambia es el gancho.
→ Dolor, "¿no tienes tiempo para el protector solar por la mañana?".
→ Curiosidad, "el paso que dejé de saltarme".
→ Recomendación de amiga, "a todas las que me preguntan qué uso les digo lo mismo".
→ Antes y después, dos pasos contra uno con SPF 30 incluido.
Para el antes y después no quiso inventarse una piel transformada, porque sería una promesa imposible de demostrar. Enseñó la rutina en lugar del resultado.
Y me avisó de no elegir ganador por clics a los pocos días, sino por compras, con al menos 10 por versión.
Quien hizo todo esto se llama Max y es el agente de anuncios que @Creatify_AI saca hoy. Si conectas tu cuenta de Meta, los lanza con tu OK y te dice cuál gana.
¿Tú cuántos ganchos pruebas por semana?
🤝 Paid partnership
Alexa Benchmark retweeted
Imagina tener un agente de IA que se encarga de las tareas aburridas por ti.
August te ayuda con el papeleo médico, desde encontrar médicos cubiertos por tu seguro y reservar citas hasta revisar tu cobertura y hacer seguimiento con los profesionales.
Me viene genial a mí, que siempre dejo las citas para más tarde por las largas esperas y el lío de organizarlo todo.
🤝 Paid partnership
We raised $12M from @Accel and @PeakXV to put AI through the bullshit American healthcare puts you through.
Navigating healthcare is hard and all of us have a list of healthcare tasks we’ve been putting off. From visiting a doctor to filing an insurance claim, the tasks pile up and get put off.
We’ve already helped 10M people understand their health. August now talks to the health system on your behalf. Tell it what you need, and August takes care of it.
In the last few weeks, August spent 24,286 minutes on hold while finding the earliest available in-network doctor appointments for our early users.
It spent 366 minutes on hold for one person alone. It also helped another save $1,255 on an insurance claim. Naturally, the ones who tried it say that they can't ever imagine going back.
I’ve held the phone for hours calling doctors, while a disease made it hard to even hold the phone. I’ve felt lost and overwhelmed when dealing with a disease and the system.
It’s why I started building August, so that all of us have a companion for our health.
Alexa Benchmark retweeted
⚡Meet Hy4 flash, the cheapest model for everyday productivity. Matches Claude Haiku 5.5 (high) on daily work (human blind tests), yet 18x task cost reduction. API pricing (per 1M tokens): $0.003 cached · $0.070 input · $0.281 output. 🔗:hy.tencent.ai/research/hy4-f…
Alexa Benchmark retweeted
Una marca de auriculares lleva unos 258 días con el mismo anuncio activo, y una IA lo encontró, me explicó por qué funciona y me hizo el mío 🎧
Quería comprobar si el nuevo agente de @Creatify_AI piensa como un media buyer de verdad o si es otra herramienta más. Así que le pedí a Max que buscara el anuncio de auriculares que más tiempo lleva activo ahora mismo.
Revisó 7 marcas en la biblioteca de anuncios y su razonamiento me convenció. La biblioteca no enseña lo que gastan, pero si un anuncio sigue activo después de 8 meses es porque está dando dinero. Y encima la marca sacó una segunda versión con otra creadora.
Luego me hizo un anuncio nuevo de 30 segundos para mis cascos con la misma estructura, sin copiar su marca, sus imágenes ni sus frases.
→ La pregunta de un amigo en el primer segundo.
→ El problema, el ruido de la cafetería.
→ La solución, se pone los cascos y el ruido desaparece.
→ El "pero lo más loco es..." con las 40 horas de batería.
Antes de entregarlo cortó un plano porque durante un segundo aparecía el logo de otra marca en un portátil.
Y si conectas tu cuenta, lo lanza en Meta o TikTok y vigila los resultados cada día, siempre con tu OK.
Espiar a la competencia y convertirlo en tu propio anuncio en una sola conversación.. ¿lo usarías?
🤝 Paid partnership
Alexa Benchmark retweeted
Deja de postular en LinkedIn en 2026.
Deja de postular en InfoJobs en 2026.
Deja de postular en Upwork en 2026.
Pasé 4 años aplicando. 1 entrevista.
Mi esposa pasó 2 horas en estos 7 sitios. 6 ofertas de trabajo remoto al instante.
Usa estas 7 plataformas para encontrar trabajos remotos ↓
Alexa Benchmark retweeted
Acabo de probar una nueva herramienta que coge tu mejor anuncio, te explica por qué funciona y te saca versiones nuevas en 1 minuto 🤯
Le enseñé un vídeo UGC de una esterilla de yoga y Max, el agente de anuncios de @Creatify_AI, me lo desmontó como un media buyer con años de experiencia.
→ Abre con el problema del cliente, no con el producto.
→ Enseña la promesa en vez de decirla, con una mano que no resbala.
→ Parece una persona real y no un anuncio.
→ Repite un único mensaje hasta el final.
Con eso me escribió un brief creativo y me sacó 2 anuncios nuevos con ganchos distintos. Hasta movió un texto que tapaba la mano antes de entregármelos.
Y esa es la mitad. Max también lanza esos anuncios en Meta, TikTok o Google, vigila tus resultados cada día y te propone cambios, pero nada se toca sin tu OK.
Resultados → análisis → brief → anuncios nuevos, todo en el mismo sitio.. ¿cuánto tiempo te ahorraría esto?
🤝 Paid partnership
Alexa Benchmark retweeted
I asked an AI to spy on my competitors' ads and it found one that's been running for almost 3 years 👀
Max from @Creatify_AI pulled live ad library data from 12 yoga and home fitness brands, plus 2,509 active US ads for "yoga mat", then picked out the long running winners.
What they have in common is almost too simple.
→ A problem or a regret in the first 2 seconds, in big captions. None of them open on a logo.
→ Filmed on a phone, vertical, one real person.
→ 15 to 40 seconds.
→ One physical benefit you can actually see, like grip.
Then it offered to watch these competitors every Monday and flag which new ads are still running after 30 days.
And that's the research side. Max also makes the ads, launches them on Meta, TikTok and Google and checks your results every day, with nothing going live without your approval.
Kind of an unfair advantage for small brands.. would you use it?
🤝 Paid partnership
Alexa Benchmark retweeted
My AI just saved me 4 days of paid time and $800 on my yearly full-body checkup.
I had a reminder that my annual checkup was due, but I didn't want to spend days calling doctor's offices and comparing prices.
So I used August. August is a healthcare agent, and it found doctors who accepted my insurance, checked which ones had appointments available, compared the estimated costs for the same tests, and helped me find a more affordable option.
i CAN SEE THIS AGENT DOING WONDERS IN AI.
🤝 Paid partnership
We raised $12M from @Accel and @PeakXV to put AI through the bullshit American healthcare puts you through.
Navigating healthcare is hard and all of us have a list of healthcare tasks we’ve been putting off. From visiting a doctor to filing an insurance claim, the tasks pile up and get put off.
We’ve already helped 10M people understand their health. August now talks to the health system on your behalf. Tell it what you need, and August takes care of it.
In the last few weeks, August spent 24,286 minutes on hold while finding the earliest available in-network doctor appointments for our early users.
It spent 366 minutes on hold for one person alone. It also helped another save $1,255 on an insurance claim. Naturally, the ones who tried it say that they can't ever imagine going back.
I’ve held the phone for hours calling doctors, while a disease made it hard to even hold the phone. I’ve felt lost and overwhelmed when dealing with a disease and the system.
It’s why I started building August, so that all of us have a companion for our health.
Alexa Benchmark retweeted
🚨 BREAKING: Okara just put an AI CMO inside Claude, Cursor and Codex, and you run it from the same chat where you code.
Ask it to fix the pages losing traffic or write a week of posts in your voice.
Paste a URL and it starts.
🤝 Paid partnership
Today, we're introducing Okara MCP.
Connect it to Claude, Cursor, Codex, or any LLM, and it finds gaps, fixes issues, and publishes your posts across socials to help you grow traffic and revenue.
Try it now at okara.ai/mcp
Alexa Benchmark retweeted
«Todo el código es basura, excepto el mío». Lauren Tan, ingeniera de GrokBot en SpaceXSI.
En el AI Engineer Code Summit acaba de mostrar toda su configuración de desarrollo.
Cada error de un agente se convierte en una regla de linting, no en un prompt. Los agentes verifican su propio trabajo con vídeos, capturas de pantalla y tests.
¿El resultado? 3.000 PRs llevadas a producción en un mes.
Su base de código está ahora tan limpia que los agentes fusionan la mayoría de sus propias PRs.
Mira esto antes de automatizar cualquier cosa.


















