GPT-6 Astra (High): benchmark results

Provider: OpenAI. Released 2026-09-03. Access: API.

Aggregate Rating 1954 ± 8, rank #17 of 2257 rated models, from 270 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AI Coding Daily (Codex CLI) - Laravel Code Quality18.6Laravel Code Quality (max 20) points, LLM-judged rubric scor100
AI Coding Daily (Codex CLI) - Total84.25Total points (max 90)100
BVB70.07Overall (%)100
Computer Anthology Terminal Tasks (Codex CLI)65.6pass@1 (%)100
Deepnote Agent Initiative Benchmark8.97Mean blind evaluator task score (0-10)100
Evolving Requirements - Evo StrictPass68.9Native score (%)100
Evolving Requirements - Evo+Comp FDAC46.5Native score (%)100
Evolving Requirements - Evo+Comp FDAC-soft52.6Native score (%)100
Evolving Requirements - Evo+Comp StrictPass72.1Native score (%)100
Evolving Requirements - Full StrictPass59Native score (%)100
FWBench0.4Decision score S (lower is better; 0 is the hindsight-optima100
Finding the Right Fit - ALE-CLI (OpenHands)53.86Task score (%), OpenHands harness, high reasoning effort, on100

Interactive version: theaggregate.ai/model?slug=gpt-6-astra-high · How It Works · Data refreshed daily, snapshot 2026-11-17.