Hard and fair benchmarks, scored live.
We build verifier-graded benchmarks with real headroom, and re-score model + agent pairs as new models ship.
Coding agents build 50 projects one feature request at a time, each checkpoint a fresh session on the code the last one left. Ranked by the share of runs whose last checkpoint passes every hidden test, three runs per task and model.
GPT 6.1 Sol 100.0% almost 83.7% 
Claude Opus 5.5 98.6% almost 75.7% 
Kimi K3 90.3% almost 60.4% Can an agent reconstruct a program from its runnable binary? Fifty held-out tasks measure partial progress and complete resolution with deterministic behavioral tests.
Claude Opus 5 50.0% almost 14.0% 
Grok 4.6 24.0% almost 6.0% GPT 5.5 28.0% almost 2.0% The set spans 17 categories and 11 languages and runtimes, weighted toward data science, ML, optimization, and model-training work, and it grades whether the artifact works within the stated constraints, not code quality or style.
GPT-6 Astra Terminus-2 · OpenAI 67.0% Claude Fable 5.1 Terminus-2 · Anthropic 62.4% 
Gemini 3.8 Flash Terminus-2 · Google DeepMind 51.8% Gullibility, measured: on tiny tasks where a convenient stored field lies and the truth sits one derivation away in the same data, does the model check or take the number it was handed? 50 held-out multi-hop tasks rendered in three substrates, scored pass@1 over five runs.
Opus 5 95% CI 39–60 49.3% Fable 5 95% CI 41–55 48.1% 
Muse Spark 1.2 95% CI 33–51 42.0% Video perception, measured: 48 curated multiple-choice questions over 26 recordings filmed to order, each built around one twist. Twelve frontier models scored pass@1 across every ordering of the options, against a five-person human baseline of 85%.
GPT-5.6 Sol 95% CI 62–87 75% GPT-5.5 95% CI 62–87 75% Kimi K3 95% CI 59–85 73% More entries are coming over the next weeks and months, each held to the same standards of deterministic verification, calibrated difficulty, uncertainty-aware reporting, and held-out tasks. Read the Computer Anthology announcement.