Stay informed about the latest AI coding model releases and benchmarks
Coverage on model launches, benchmark results, and ecosystem changes to help you follow how coding assistants evolve.
A practical guide to understanding SWE-bench benchmark families, scaffold effects, and reproducibility — and how engineering teams should evaluate AI coding agents beyond a single leaderboard snapshot.
Anthropic confirmed an internal Claude Code source package was exposed due to a release packaging mistake. We break down confirmed facts, what rival teams can learn, and what it means for engineering teams evaluating AI coding agents.
Use a practical scorecard to evaluate AI coding agents by acceptance rate, reviewer time, regression risk, and cycle time—not benchmark hype.
OpenAI, Anthropic, and others are shipping agentic coding updates fast. Use this practical framework to compare coding models by delivery speed, review burden, and production reliability.
Don't pick coding agents by leaderboard alone. Use this 2026 evaluation framework with repo-fit tasks, review burden, failure rates, and total delivery speed.
Google introduces Finish Changes and Outlines in Gemini Code Assist extensions, giving developers structured control over AI-generated edits in IntelliJ and VS Code.
Anthropic and OpenAI both release major coding model upgrades — Claude Opus 4.6 and GPT-5.3 Codex — intensifying the AI coding arms race.
Anthropic releases Claude Opus 4.5, delivering state-of-the-art software engineering performance with significant improvements in agentic tasks, multi-step reasoning, and autonomous coding capabilities.
Anthropic unveils Claude Sonnet 4.5, claiming the title of best coding model in the world with state-of-the-art performance on SWE-bench and significant improvements in reasoning, math, and agent capabilities.
Anthropic is unifying all developer offerings under the Claude brand, streamlining naming across the platform while keeping all technical implementations unchanged.
OpenAI unveils GPT-5 Codex, a specialized version of GPT-5 optimized for autonomous coding tasks, featuring enhanced code review capabilities and seamless integration across development environments.
DeepSeek has released version 3.1 of their AI model with significant improvements in code generation and reasoning capabilities.
Our comprehensive benchmark reveals how Google's Gemini 2.5 Flash balances lightning-fast performance with code quality.
We tested Claude 3.5 Sonnet against GPT-4 across 100+ coding challenges to determine which AI truly codes best.