Inference

deepseek-v4.1-flash-1017-cyber is now available on @inference_net. This is the first of several highly capable cyber models we'll release over the coming weeks. We believe it is our duty as technologists to help people protect themselves and their digital properties from the current onslaught of cybercrime ravaging much of the world. As such, we’re serving our cyber models “at cost”, in US data centers only, with ZDR enabled by default. KYC required. Apply here

RT @samhogan: 8th day in a row of > 1 trillion tokens on OpenRouter:) lots of new stuff coming soon™️

RT @samhogan: If you’re the founder of a company operating at reasonable scale and you’re not actively planning to spend >$100k to try to hack yourself in the next 30 days using agent swarms I beg you to reconsider DMs open if you want to chat shit is crazy. stay safe out there fam

Some game development: "Make an interactive 3d game about a first person pov electrical technician that has to work in a datacenter and you have to move around and plug in GPUs, get the power offline, fix issues that pop up like GPUs overheating, technical issues, and whatnot. You are basically a technician that slowly gets capacity online and you win when the full datacenter is occupied. make it have several interactions and technical details where you have to overcome challenges, can buy new equipment, can progress through the game, some funny interactions, and overall just a fun gaming experience. Add as many details as you can to make this a fun game."

and of course the classic "make a svg of a pelican riding a bike"

We've hosted the model with some competitive pricing. If you want to try it out, head over to our site at

🚀 We're excited to be an official Day 0 Provider for @MiniMax_AI M3.1 Flash! Built for coding, agents, and complex workflows, M3.1 Flash brings powerful capabilities at a competitive cost. We tried it out on a few prompts...

Starting off basic... "Build an interactive double-pendulum laboratory that makes chaos feel tangible. Show two pendulums starting almost identically, then diverging. Let me change their masses, lengths, starting angles and tiny initial difference, pause time, step the simulation and inspect their trails, energy and phase portraits. Make it feel like a beautiful scientific instrument, with an intuitive explanation of why deterministic does not mean predictable."

RT @samhogan: inference

Fast tokens, Faster reflexes 🏓 We’re teaming up with @depotdev for a ping pong tournament. 3 tables, 1 champion, and a Nintendo Switch 2 on the line. Oct 27 at our office by the Embarcadero. Come swing by!

October 1st 🚀 we’re taking over a sick theater for a private screening of Interstellar 🚌 Coach bus leaves the Inference office @ Embarcadero at 5pm 🍻 Come pregame with us beforehand 🍴 Catered food + beers + Inference merch at the theater 🚌 Back at the Inference office by 9:30pm Come hang , spots are limited make sure to RSVP :)

RT @atbeme: excited to share that @inference_net is officially live on the @vercel ai gateway!

RT @samhogan: introducing fast inference ( fast inference is an LLM API for devs who want to go faster access the fastest Kimi K3, GLM, DeepSeek, etc for $9/mo integrations with Claude Code, Codex, OpenCode & more speeds compounds. waiting for Claude does not

RT @samhogan: I’m running an experiment with a handful of folks who spend > $5k/mo on coding models to better understand the state of open source models for complex coding tasks. If you’re interested in driving Kimi K3 for a few days and giving feedback, shoot me a DM Paying $500 each

RT @TheHarryET: It’s crazy how much a 3x speed difference makes for coding agents. here's a side-by-side of both models re-creating the github home page. glm is halfway done before sonnet even starts. glm 5.2 fast (@inference_net) = ~300 tps sonnet 4.5 (via anthropic) = ~100tps even crazier: glm 5.2 doesn’t have image support and still works for this

RT @samhogan: We're releasing Inference AutoEvals⚡️ automatically find the best model for any agent by replaying traffic across a suite of models. works with every model/provider on average, AutoEvals cuts spend by 30% & improves accuracy by ~10% with a one-line code change Available now👇

RT @samhogan: Want to try Kimi K3 in production but worried how it might change your product? Don’t worry, we got you: 1. Install Inference Gateway ( 2. Keep sending traffic to your current provider 3. Gateway automatically starts sorting through your live data using an RLM to generate evals for your app. This takes ~24 hours. 4. Gateway starts mirroring live traffic to Kimi K3 to run evals. Traffic is mirrored - you’re still using your old model in prod. 5. Once evals look healthy, you get a Slack notification letting you know it’s safe to switch. 6. Switch model id in your code to “kimi-k3” Congrats, you just saved 50% on your monthly token bill, and you own your LLM stack end to end.

RT @samhogan: We’re releasing Inference AutoRouter. Always the right model for your task AutoRouter evals dozens of models and routes requests based on cost/accuracy One endpoint. Hundreds of models Append `:auto` to the model name to enable AutoRouter Available in private beta today 👇

RT @julesrosenberg: 5 things @samhogan does differently > signs company documents from the terminal > trained 5 function-specific @inference_net agents > keeps his entire to-do list in one apple note (over a year old) > moved his entire team to @opencode for llm portability > blocks twitter on his phone from 9-5 (300+ muted words) ep 6 of show me your stack is live!