Agents Last Exam
Best computer-use accuracy at lower API cost
Astra reaches 59.3% accuracy, compared with 53.6% for GPT-5.6 Sol and a reported 55.5% for Claude Opus 5. At peak settings, Astra uses approximately 65% fewer output tokens than Opus 5.
Agents’ Last Exam
Accuracy
API cost per task (USD)
Opus 5 is shown as a reported-score reference; no API-cost point is provided. View OpenAI source
