Tutorials, deep dives and product notes — built for developers.
Cognition’s FrontierCode 2.0 is a new 132-task mergeability benchmark. Interactive leaderboard with best-effort model scores, task pass rates, flag rates, harnesses, and actual evaluation cost per rollout. Updated October 30, 2026.
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.
Archived FrontierCode 1.1 Main leaderboard: Cognition results remain separate from independent Artificial Analysis runs; adds Kimi K3.1 Pro’s 47.1% AA Stirrup result. FrontierCode 2.0 is current and not score-comparable. Updated Oct 30, 2026.
DeepSWE v1.1: adds Kimi K3.1 Pro (75.5%) and MiniMax M3.1 Flash (73.7%) as provider/model-card runs, separate from Datacurve's mini-swe-agent board. Muse Wonder Full (79.0%) remains the top provider score. Updated Oct 30, 2026.
Terminal-Bench 2.1: adds Meta Muse Spark Open at 76.4% from Meta’s Oct 15 model card, labeled separately from tbench.ai public rankings. DeepSeek V4.1 Flash remains public-board leader at 90.6%. Updated Oct 23, 2026.
SWE-bench Pro: Anthropic reports Opus 5.5 at 89.9%, Sonnet 5.5 at 81.3%, and newly added Haiku 5.5 at 64.8%. Vendor results remain separate from Scale's standardized rows. Updated October 16, 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.