Kimi K3 Sets Open-Weight Record as OpenAI and Anthropic Battle for Coding Dominance

·2 views

Moonshot AI released Kimi K3 this week, a 2.8 trillion-parameter multimodal model with a 1 million token context window, with open weights expected by July 27; it is already live on their API and reportedly outperforms several top closed models in coding benchmarks. Separately, Thinking Machines Lab, founded by former OpenAI executive Mira Murati, launched Inkling, a 975 billion-parameter open-weights foundation model trained on 45 trillion tokens across text, images, and audio. OpenAI consolidated its Codex tool into ChatGPT under the rebranded product ChatGPT Work, which saw 2.5x usage growth in a week, though the move drew backlash from power users whose established workflows were disrupted. Anthropic responded swiftly by resetting usage allowances for all Claude users, signaling an intensifying competition to become the default coding agent for developer teams. Industry observers note that the broader race is shifting from raw model capability toward the infrastructure and tooling ecosystems built around AI models.

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)

How AI Companies Hijack the Gartner Hype Cycle to Avoid Accountability

The Gartner Hype Cycle describes the typical arc of a new technology, moving from initial excitement through inflated expectations, disillusionment, and finally stable productivity. Critics argue that AI companies deliberately disrupt this natural progression by releasing faster or cheaper model versions — branded as breakthroughs — just before a product matures into a commodity. This strategy resets the hype loop at Stage 4, allowing firms to restart overpromising without being held accountable for unresolved limitations like hallucinations, high costs, or missing features. The tactic also targets non-technical executives focused on cost-cutting rather than engineers who evaluate reliability. The result, according to this analysis, is that genuine technological progress is replaced by a repetitive marketing cycle that keeps users chasing promises rather than benefiting from stable, proven tools.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

1C-Bitrix online store load time cut from 15s to 2s with SQL and caching fixes

A Russian federal IT equipment distributor was experiencing severe performance issues, with catalog pages taking 7–15 seconds to load and a bounce rate of 62%. Over three working days, a technical team diagnosed the root causes by analyzing server logs rather than immediately adjusting cache settings. They identified three core problems: slow SQL queries on large catalog tables, absent Nginx-level caching that routed every request through PHP, and uncompressed images inflating page size to 4.2 MB. Fixes included adding database indexes, enabling FastCGI and Bitrix component caching, converting images to WebP with lazy loading, and minifying CSS and JS files. The result was an LCP of 2 seconds, a page size reduction from 4.2 MB to 1.1 MB, a drop in bounce rate to 38%, and a search ranking improvement of 15–20 positions — all without upgrading the server.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Engineering Team Cuts RAG Pipeline Latency 40% Using Bayesian Search and Hybrid Retrieval

A development team overhauled their Retrieval-Augmented Generation (RAG) pipeline after standard production deployments revealed significant performance gaps with legal, API, and support ticket documents. The team replaced fixed 512-token chunking with document-type-specific strategies, including recursive, semantic, and agentic chunking, achieving recall@10 scores between 91% and 97% across content types. They also implemented a hybrid retrieval system combining vector search and BM25 keyword search, fused via Reciprocal Rank Fusion and re-ranked using a cross-encoder model. The cross-encoder reranking step added only 50ms of latency while improving recall by 15%, contributing to an overall 40% reduction in end-to-end query latency. The team noted that query transformation and parallel retrieval stages were equally critical to reaching production-grade performance.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Fix Puppeteer Memory Leaks in Production Before They Crash Your Container

Puppeteer memory leaks in production are rarely a single bug — they typically stem from unclosed pages, long-lived browser instances, and orphaned Chromium processes left behind after crashes or timeouts. Developers often misdiagnose the issue by monitoring Node.js heap usage, while the actual memory growth occurs in Chromium's child processes outside the Node process. Key fixes include always closing pages and browser contexts in a finally block, recycling the browser after a set number of jobs rather than keeping one instance alive indefinitely, and running Chromium under an init process to reap zombie processes. Concurrency should be capped and memory ceilings set to limit the damage a single misbehaving page can cause. For use cases limited to capturing screenshots of URLs, a hosted screenshot API can eliminate the need to manage Chromium entirely, removing the memory problem at its source.

0 comments Read more at DEV Community