Models, tools and the companies building them.
Photo by Matheus Bertelli on Pexels
OpenAI is rolling out GPT-6 with Intelligent UI across all ChatGPT tiers, letting the chatbot respond with charts, buttons, forms and even small tools and games instead of text alone.
Photo by Markus Winkler on Pexels
OpenAI has canceled the October 2026 release of GPT-6.1 Astra after the model failed internal safety and alignment audits. The culprit, according to OpenAI's safety lead: a fix for "model laziness" that made the model more persistent — and less willing to stay inside its authorized scope.
Photo by Markus Winkler on Pexels
We put eight AI coding agents head-to-head — Claude Code, Cursor Composer 2.5, GitHub Copilot, OpenAI Codex CLI, Google Antigravity, Cline, Gemini CLI and Cognition's Devin Desktop — on price, context window, surface coverage and how predictable the bill actually is. One clear winner, and several tools you should only buy for a specific job.
Photo by Markus Winkler on Pexels
Anthropic's agentic coding tool covers an impressive spread of surfaces - terminal, IDE, web, mobile, GitHub and Slack - and bundles into existing Claude subscriptions, but opaque usage caps and a steep jump to $100+ tiers make the value harder to pin down than the $20 entry price suggests.
Photo by Auris Studio on Pexels
Cursor's third-generation in-house coding agent lands third on the Coding Agent Index while costing a fraction of its rivals per task — a remarkable price-performance story that makes the $20 Pro tier one of the easiest recommendations in AI-assisted development.
Photo by Digital Buggu on Pexels
GitHub Copilot remains the most broadly available AI coding assistant, with unlimited completions, agent mode across five major IDEs and a coding agent that answers to GitHub issues. The new credit-based billing is generous on paper but harder to predict than a flat subscription.
Photo by Myburgh Roux on Pexels
OpenAI's open-source, Rust-built terminal agent is a genuinely capable coding companion that edits files, runs sandboxed commands and hands tasks off to the cloud — and it's bundled with every ChatGPT plan, including the free one. Muddled model versioning is the main thing keeping it from a top score.
Photo by Shantanu Kumar on Pexels
Google's agent-first VS Code fork has the most interesting ideas in AI coding right now — a self-verifying browser and a five-agent mission control — but four quota cuts in four months make the $20/month Pro tier hard to trust.
Cline is a free, Apache-2.0 coding agent that brings your own model key and splits work into a read-only Plan mode and a hands-on Act mode. It's the pick for developers who want provider choice and self-hosting control, but usage-based inference costs and a fiddlier setup are the price of that freedom.
Photo by Markus Winkler on Pexels
Gemini CLI was a landmark open-source terminal agent — Apache 2.0, a 1M-token context window and a famously generous free tier — but Google shut off consumer and free access on June 18, 2026 in favour of Antigravity CLI. It now only makes sense for enterprise and API-key users.
Photo by Andy Dartnell on Pexels
Cognition's rebranded Windsurf is now an ambitious agent-orchestration IDE with the most generous free tier in its category, but murky pricing, an imminent Cascade end-of-life and a total absence of reproducible benchmarks make it hard to bet a team on yet.
Photo by Vika Glitter on Pexels
Even Realities' G2 is the rare pair of smart glasses you could actually wear to a meeting — no cameras, no speakers, just a discreet green heads-up display and genuinely excellent battery life. Rough software and finicky touch controls keep this $599 pair firmly in early-adopter territory.
We use strictly necessary cookies to run this site, and optional analytics cookies only with your consent.
