Moonshot AI releases Kimi K3, the largest open-weight AI model at 2.8 trillion parameters

·9 views

Chinese AI firm Moonshot AI released Kimi K3 on July 17, 2026, making its weights publicly downloadable on July 27. With 2.8 trillion parameters, it is the largest open-weight model ever published and benchmarks suggest it rivals leading closed models like Claude Opus and GPT in coding, reasoning, and general knowledge tasks. Despite being freely downloadable, the model requires an enterprise-grade GPU cluster to run, putting it well beyond the reach of hobbyists or home lab users. The release reignites debate over AI openness and geopolitics, echoing tensions that surfaced when DeepSeek-R1 launched in early 2025. While K3 itself is not consumer-runnable, its techniques are expected to filter down into smaller distilled models that could appear in platforms like Ollama within weeks.

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)

Kiro Crew AI Agent Diagnosed a Simulated Payment Outage in 33 Seconds

An engineer tested Kiro Crew, an open-source AI agent orchestrator, by feeding it a simulated P1 incident involving a 22x payment service latency spike. Within 33 seconds, the agent ran seven parallel diagnostic steps, correlating git history, cron schedules, and system resource data to form a root cause hypothesis. It flagged a recent dependency bump to pg-pool and a connection timeout change as likely culprits behind connection pool exhaustion. The same investigation would typically take a human on-call engineer 15 to 30 minutes. Beyond incident response, the agent was also used to schedule automated Monday morning infrastructure reports, replacing a manual weekly task the engineer had performed across multiple client sites.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Tools Write Faster Code Than Juniors, But Judgment Gap Remains Wide

A senior developer writing for DEV Community argues that GitHub Copilot can complete in seconds what once took junior developers hours, citing Stack Overflow survey data showing 84% of developers now use or plan to use AI tools. In side-by-side comparisons, AI-generated code tends to be syntactically cleaner and better commented, but frequently misses undocumented business context or solves the stated problem rather than the underlying one. The author contends that junior roles are not disappearing but transforming, with foundational skills shifting from producing code to critically evaluating AI output. Developers at all levels are advised to treat AI like a confident but overconfident intern — useful for speed, but requiring experienced oversight. The piece concludes that skills like systems thinking, architectural judgment, and mentorship remain distinctly human strengths that AI tooling has yet to replicate.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Beginner's Guide: How to Intercept Web Traffic Using Burp Suite on Windows

A step-by-step tutorial explains how to set up Burp Suite Community Edition on Windows 11 Pro to intercept and inspect live web traffic. Users are guided through launching the tool, enabling the proxy listener on 127.0.0.1:8080, and using Burp's built-in browser to avoid manual certificate configuration. With Intercept turned on, visiting a test site causes the HTTP request to pause in the Intercept tab, where it can be reviewed and forwarded. Forwarded requests are then logged in the HTTP History section with full request and response details. The guide also addresses a common beginner issue where browser caching prevents requests from being intercepted, offering fixes such as disabling cache via DevTools or performing a hard reload.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Indie Game Moksha Adds Tutorial Button to Start Screen in Devlog 9

Developer of the browser-based game Moksha has addressed a UX gap by adding a dedicated tutorial button directly on the start screen in commit f0ae956. Previously, players who missed the 8-step Sanskrit scripture-style tutorial had no UI option to replay it, as the only trigger was the in-game 'R' key. The new button clears the localStorage key 'moksha_tutorial_seen' and relaunches the tutorial fresh, mirroring the readiness-gating behavior of the main start button so it only activates after audio and fonts fully load. The update also fixes two variable-name typos that had caused timer reset and canvas-click dismiss failures. The button supports the game's Hindi/English language toggle and uses a secondary outlined visual style to maintain clear UI hierarchy beneath the primary start button.

0 comments Read more at DEV Community