Tutorials, deep dives and product notes — built for developers.
Interactive Terminal-Bench 3.0 leaderboard with Claude Opus 5 at 42.7%, GPT-5.6 Sol at 34.6%, GLM 5.3 at 28.3%, and all tracked frontier agent models ranked. Updated August 2026.
FrontierCode 1.1 Main: Opus 5.5 leads at 54.6%; new Cognition entries GPT-6.1 Sol (50.2%) and Claude Haiku 5.5 (46.4%) include effort, rollout spend, API rates and token details. Updated October 9, 2026.
DeepSWE v1.1: Datacurve's standardized leaderboard is kept separate from provider runs. Adds Google-reported Gemini 4 Argon at 77.9% and Ornith-1.5-397B at 56.0%; per-task costs are marked unavailable. Updated October 9, 2026.
Terminal-Bench 2.1 leaderboard with DeepSeek V4.1 Flash atop the public board and newly tracked Ornith-1.5-397B at 86.1% on Terminus-2 as a separate vendor run. Vals AI and Cognition runs remain distinct. Updated October 9, 2026.
SWE-bench Pro: Anthropic-reported Opus 5.5 leads at 89.9%; newly tracked vendor results include Ornith-1.5-397B (65.1%) and Qwen3.8-Omni-Flash (63.3%), kept distinct from Scale standardized results. Updated October 9, 2026.
What SWE-bench Pro actually measures, how it works (1,865 tasks, 41 repos, 123 languages), why OpenAI abandoned SWE-bench Verified, the DeepSWE audit that found 32% verifier errors, and how to use coding benchmarks correctly. The definitive explainer.