SShortSingh.
Back to feed

Moonshot AI Releases Kimi K3, the Largest Open-Weight AI Model at 2.8T Parameters

0
·1 views

Beijing-based Moonshot AI, backed by Alibaba, released the full weights of its Kimi K3 model on Hugging Face on July 27, 2025, making it the largest open-weight AI model ever published at 2.8 trillion parameters. K3 uses a mixture-of-experts architecture that activates only a fraction of its parameters per token, making it feasible to run despite its massive size. The model tops independent frontend coding benchmarks, outperforming closed flagship models from Anthropic and OpenAI, and supports a context window of over one million tokens with flat pricing at $15 per million output tokens. However, independent testing by developer Simon Willison revealed that K3's reasoning-heavy token usage can drive up real-world costs significantly, with one simple task consuming over 13,000 reasoning tokens before producing any output. This gap between benchmark performance and practical cost efficiency raises questions about how well the model's marketing aligns with actual deployment economics.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

How to Build Fingerprint Verification in ASP.NET Core Using Open-Source SourceAFIS

A developer tutorial published on DEV Community explains how to implement 1:1 fingerprint verification in ASP.NET Core using SourceAFIS, a free open-source library. The guide covers the full pipeline: decoding an uploaded fingerprint image into raw grayscale pixels, extracting a minutiae-based template, comparing two templates to produce a similarity score, and applying a threshold to determine a match. The article distinguishes 1:1 verification — confirming a fingerprint matches a specific enrolled user — from the more complex 1:N identification problem of searching across an entire database. SourceAFIS handles the core matching mathematics, while the developer is responsible for setting the threshold and building the API plumbing. The tutorial uses SixLabors.ImageSharp for image decoding and wraps the logic in a minimal ASP.NET Core web API, noting that a production-grade system would require additional considerations beyond this basic implementation.

0
ProgrammingDEV Community ·

Guide: Deploy a Python Flask App on a $5/Month DigitalOcean VPS

A developer tutorial outlines how to host a Python web application on a budget-friendly virtual private server using DigitalOcean's entry-level $5/month plan, which provides 1 CPU core, 1 GB of RAM, and 30 GB of storage. The guide walks through provisioning a Ubuntu 20.04 droplet, setting up a Python virtual environment, and building a basic Flask web application. It also covers installing Nginx as a reverse proxy to serve the Flask app to the public internet. The tutorial is aimed at Python developers looking for an affordable and straightforward path to deploying their projects beyond local development.

0
ProgrammingDEV Community ·

5 Interview Questions Non-Technical Founders Should Ask First-Time Dev Hires

Hiring a first developer is one of the riskiest decisions a non-technical founder makes, largely because they lack the tools to properly evaluate a candidate's actual competence. A common failure mode is allowing candidates to dominate interviews with technical jargon, which rewards confident performance over real ability. The article argues that useful interview questions must produce signals a non-technical founder can interpret without needing deep technical knowledge. Rather than relying on algorithm puzzles or outsourcing evaluation to a developer friend, founders are advised to ask questions that reveal judgment, ownership, and trade-off thinking. Bringing in a fractional CTO or experienced technical advisor during the hiring process is recommended to help read signals the founder cannot assess alone.

0
ProgrammingDEV Community ·

Developer Launches DocPilot 2.0, an All-in-One AI Workspace for Job Seekers

A developer has launched DocPilot 2.0, an AI-powered career workspace designed to consolidate the entire job application process into a single platform. The tool evolved from a basic AI resume builder after its creator recognized that modern job searches require far more than just a polished resume. DocPilot 2.0 includes features such as ATS optimization, a job-description matcher, cover letter generation, LinkedIn profile improvement, and an interview coaching tool. A Career Vault stores users' professional details centrally, allowing all AI features to reference the same information without repeated data entry. The platform also offers an application tracker and support for multiple resume versions, aiming to replace the fragmented mix of tools most job seekers currently rely on.