Pick the model and the stack for the job

Models, frameworks, prompts, and tools for AI coding, arranged by the job you are doing. Every recommendation has a source and a date, and says when not to use it.

Latest changes

10-11BoardDeepSWE v1.1: 28 Best-config rows. GPT 6 Astra 74.1%, Gemini 3.8 Flash 73.8%, Claude Opus 5 73.6%. General board on Web frontend / Android / CLI.deepswe.datacurve.ai

10-11BoardWrite docs & copy: new scenario on The Slop Index, 39 rows. Source ranks sloppiest first; after inversion GPT-5.6 Luna leads (100 − Slop = 74).theslopindex.com

10-09BoardLMArena Image-to-WebDev: 64 models. Claude Opus 5.5 (max) scores 1749. This is image-to-page, not the text-prompt WebDev board.arena.ai

10-09Boardsudo L7: Claude Opus 5.5 and Claude Sonnet 5.5 both score 45%, with Opus listed first. GPT 6 Astra scores 32.2%.surgehq.ai

10-08BoardLMArena WebDev: Claude Opus 5.5 (max) leads at 1813, 27 points ahead of GPT-6 Astra.lmarena.ai

09-29BoardTerminal-Bench 4.0: GPT-6.1 Sol ties for 3rd at 58.2%, at $634.24.tbench.ai

09-29PickThe full-stack set uses GPT-6.1 Sol from Terminal-Bench: best once cost is included.编辑

09-22BoardTerminal-Bench 4.0: Claude Opus 5.5 takes first at 64.8%.tbench.ai

08-11BoardAndroid Bench: adds eight models, including Claude Opus 5, GPT 5.6 Sol / Luna / Terra, Kimi K3, and Qwen3.8 Max.android.com

—MethodSWE-bench Verified has not listed a new model generation since February 2026, so it is not primary evidence.编辑

All changes RSS